How Textual Criticism Works

"Knowledge of documents should precede final judgement upon readings." -- B. F. Westcott and F. J. A. Hort, 1882¹

"witnesses are good because of their good variants, variants are good because of their good witnesses. This circle cannot be avoided, but it has to be controlled." -- Gerd Mink, 2004²

Summary

When the handwritten copies of a New Testament book disagree, an editor has to decide which wording the author most likely wrote, or at least which wording the surviving copies started from.  This article explains how that decision is made and tests the main schools against seven real places where the copies disagree.  The manuscript counts and the famous long passages are in the sibling article New Testament Manuscripts & Variants.

1. The target is a wording, not a truth claim.  The editors of the main critical edition aim at the "initial text," the form from which the surviving copying started.  Gerd Mink, who designed their method, says it "is not identical with the original, the text of the author."²  The same editors also call their reconstruction of the Catholic Letters "a hypothesis about the texts of the authors."³  Recovering a wording tells you what a book said.  It does not tell you whether the book is true, who wrote it, or whether it belongs in the canon.

2. Copies are evidence about ancestors, so they are not votes.  A thousand copies of one lost exemplar testify to one exemplar.  Westcott and Hort made this a founding rule.¹  The leading defender of the Byzantine text, Maurice Robinson, accepts it too:  "the raw number of MSS should be intelligently reduced."⁴

3. Byzantine priority is not majority voting, and the Textus Receptus is neither the Byzantine text nor one fixed text.  Robinson calls simple "nose-counting" a caricature of his position, and he says "the Byzantine Textform is not the TR."⁴⁴  His edition counts "over 1800" places where the printed Textus Receptus differs from the Byzantine text.⁴  The Trinitarian Bible Society, which publishes the Textus Receptus, concedes:  "There is no single Received Text."⁵

4. No single rule wins every case.  At Romans 5:1 the critical text prints the Byzantine reading against the first hands of its two oldest codices.  At 1 Peter 4:16 it changed in 2012 to the Byzantine reading against the papyrus and the great codices.  At Mark 1:2 it rejects the Byzantine reading because "Isaiah" has older and more widely spread support.  The shorter-reading rule fails for the earliest scribes whose habits were measured.  Of the seven places, the Byzantine and critical texts differ at three, Mark 1:2, John 1:18 and 2 Peter 3:10, and only Mark 1:2 is settled with fairly high confidence.

5. The genealogical computer method (CBGM) tests editors' choices;  it does not make them.  Its users write that "the CBGM is not making the decision for us."³  It relates texts, not physical books, and its family relations rest on the editors' own judgments about which reading came first.³  At 1 Peter 4:16 the relative closeness of the witnesses supports the editors' view that "in this name" arose more than once, under their chosen settings.  Broadening the search for ancestors weakens that result, and Vaticanus's particular link restates the local choice.  Part 4 shows the data and their limits.

6. The majority argument, priced exactly.  Zane and David Hodges claimed that if 500 copies split 75:25, the chance the minority holds the original is "less than one in ten" under their copying model.⁶  Computed exactly, that is true at low error rates and false at the one-in-three error rate they used.  At that rate a lopsided split is extremely rare;  given one, the minority holds the original in roughly 31 to 50 percent of cases, depending on which generations survive.⁷  Their further argument from many majority readings requires attention to shared copying and survival.  Preferential copying of one ancestor carrying many variants could raise them together.  Whether that happened to the Byzantine text is a historical question counting cannot answer.

7. Preservation is a belief;  transmission is a record.  The Westminster Confession teaches that Scripture was "by his singular care and providence, kept pure in all ages."⁸  A belief stated that generally fits any manuscript record.  Specific forms of it make claims that can be checked.  Edward Hills held that providence worked "publicly in the usage of God's Church," so that the true text is "found in the majority of the New Testament manuscripts" and the Textus Receptus was "God-guided."⁹  At Revelation 22:19 the Textus Receptus reads "book of life."  The current apparatus finds that Greek reading in a sixteenth-century manuscript and an undated correction to a fifteenth-century one.  Neither establishes a Greek reading before Erasmus's 1516 edition;  the manuscript Hills offered as Erasmus's possible source reads "tree."⁹¹⁰¹¹

Verdict.  The schools agree more than their reputations suggest:  all reduce copies by descent where descent can be shown, and all weigh internal evidence.  On the cases worked here, the critical editions give reasons that can be checked, mark their doubts, and have moved toward the Byzantine text where their evidence pointed there.  Byzantine priority is a coherent rival, and the places where it and the critical text disagree remain open or only moderately settled.  Selected test cases cannot rank the methods across the whole New Testament.  Textus Receptus positions rest first on a theological premise.  Where their defenders make specific factual claims, some fail on the evidence, Revelation 22:19 most clearly, and others are matters of degree.  The uncertainty that can be measured sits at known places among the surviving witnesses.  In the Catholic Letters, for example, the editors decline to choose at 43 places and print one reading that no Greek copy has.³³  Changes made before the earliest witnesses, if they left no trace, cannot be measured by any method.

Part 1: What the editor is trying to recover

Four possible targets

An editor can aim at four different things, and much confusion comes from mixing them up.

1. The autograph. The wording the author wrote or dictated, or the first copy he approved. No autograph of any New Testament book is known to survive.
2. The initial text. Mink, who coined the term, defines it as "the reconstructed form of the text from which the manuscript transmission started." Tommy Wasserman and Peter Gurry, who wrote the standard introduction to the method, draw the consequence: "the initial text may refer to the author's text or to something later."³ Mink adds that between the autograph and the initial text "considerable changes may have taken place which may not have left a single trace in the surviving textual tradition."² The editors of the Catholic Letters take the simpler working assumption. They write that "the present reconstruction is a hypothesis about the texts of the authors," and Wasserman and Gurry read them as saying they "see no reason to believe there is any great difference" between the two.³ The distinction is a caution about what can be known. It is not a finding that the two texts differ at any particular place.
3. An authorial revision or a second edition. Some books may have circulated in two forms from the start. This is a live question for Jeremiah, whose Greek translation is markedly shorter than the Hebrew, and it is touched on in Which Bible Text Gives the Right Date for Creation? rather than here.
4. A received edition. A printed text that a church or tradition has adopted, such as the Textus Receptus behind the King James Version. Here the question is not "what did Paul write?" but "what did this edition print?", and it is answered by reading the edition.

The schools aim at different targets.  A reasoned eclectic aims at the initial text and treats it, unless there is evidence otherwise, as the best estimate of the autograph.  Robinson aims at the autograph and holds that the Byzantine text, read through a history of transmission, reaches it.⁴  Textus Receptus defenders such as Hills add a theological claim that providence guided the printed received text.⁹  The first two are claims about history that manuscripts bear on.  The third is examined in Part 7.

What recovering a wording does not decide

Suppose every word of Mark were recovered with certainty.  That would show what Mark wrote.  It would not show that what he wrote happened, that he is the Mark of church tradition, or that the book belongs in the Bible.  Those are separate questions with separate evidence, treated in their own articles, including The Biblical Canon and Inerrancy and the Missing Autographs.  The relation runs in a limited way in both directions.  Where a passage's wording is uncertain, so is the content of that report, and a historian should lower confidence in whatever rests on the disputed words alone.  But an uncertain wording does not make the events false, and a perfectly preserved text can still be fiction.  A well-preserved text shows what was claimed, not that the claim is true.

Why the question is answerable at all

Copying leaves fingerprints.  When a scribe makes a change, the copies made from his copy usually inherit it, unless a later scribe corrects it against another exemplar.  So shared changes point to shared descent, much as shared mutations do in genetics.  Westcott and Hort put the principle in one line:  "identity of reading implies identity of origin."  They added the caveat at once:  "Strictly speaking it implies either identity of origin or accidental coincidence, no third alternative being possible."¹  Scribes sometimes make the same easy change independently, and scribes who compare exemplars mix lines that were separate.  Telling coincidence and mixture from common descent is much of the craft.

Part 2: External evidence: who says what, and how independent they are

Three kinds of witness

1. Greek manuscripts. Papyri, parchment books in capital letters, later books in minuscule script, and lectionaries arranged for church reading. They give the actual Greek words, including small differences of word order and spelling that a translation hides.
2. Early translations. Latin, Syriac and Coptic versions were made from Greek copies older than most surviving Greek ones. They show which readings existed where and when. They cannot always show the Greek behind them. Latin, for example, has no definite article, so a Latin copy cannot show whether its Greek source had one.
3. Quotations in early Christian writers. These can be dated and placed. They are also often loose, quoted from memory, or adjusted to the writer's sentence, and later scribes sometimes "corrected" a father's quotation to the text they knew. A father who comments on the wording itself, or who says what copies read, is the most useful kind.

Different counts, different denominators

The numbers in this debate count different things.  The count of manuscripts, about 5,300 for the Greek New Testament, is discussed in the sibling article.  The count of variants counts every difference in every manuscript, so it grows with the number of manuscripts.  A third number is the count of places of variation:  the points in the text where the copies differ at all.  In the seven Catholic Letters, the editors of the main critical edition collated 123 selected witnesses and found 3,046 such places.¹²  Agreement percentages use that third denominator.  Codex Vaticanus (03) and Codex Sinaiticus (01), both from the fourth century, are both present at 2,999 of those places and agree at 2,613, or 87.1 percent.  The average for pairs of witnesses is 87.6 percent, the highest pair 99.1 and the lowest 77.9.¹²

These figures describe raw similarity at selected places in seven short letters.  They do not measure whether two manuscripts share an edited ancestor.  Two copies can descend from one edition and still differ often, because each line kept changing after the split.  So the slightly-below-average agreement of the two codices neither proves nor refutes a theory that an early Alexandrian edition stands behind both.

Age: the age of the parchment and the age of the text

An old manuscript shows that a reading existed by the time it was copied.  It does not show that the reading is original.  Robinson states this well:  "While older MSS, versions, and fathers demonstrate a terminus a quo for a given reading, their respective dates do not confer authenticity; they only establish the existence of a given reading at a given date."⁴  Reasoned eclectics agree in principle.  The method's designers distinguish "the age of the text and the age of the ink and parchment," because "it may well be that a young manuscript contains a text antecedent to a much older manuscript."¹²

Two facts show the point in both directions.  Manuscript 1739 is a tenth-century copy, yet a note its scribe copied from his exemplar says the text came from a "very ancient exemplar" with a text tied to Origen.  Ehrman concludes it is "the conscientious transcription of a fourth-century exemplar."¹³  A late book can carry an early text.  The reverse also holds.  An early book can carry many slips:  the scribe of the third-century papyrus P45 "omits more than twice as often as he adds."¹⁴  That is a measure of one kind of error in one copy, not a score for the copy as a whole.

The dates themselves are estimates.  The official register in Münster lists P66 and P75 as early third century, 200 to 225.¹¹  Brent Nongbri argued in 2016 that for P75 "the first half of the fourth century should certainly be regarded as a possible date."¹⁵  Arguments in this article that lean on a papyrus date say which date they use and what changes if the later date is right.

A reading can also be shown to be old without any old manuscript.  Robinson argues that the Byzantine text "reaches at least as far back as its predecessor exemplars of the late fourth and early fifth century, as reflected in MSS A/02 and W/032."⁴  Ehrman argues that the Byzantine reading at John 1:18 must be early because, he reports, Irenaeus, Clement and Tertullian used it "before our earliest surviving manuscripts were produced."¹³  Both arguments infer a lost ancestor from its descendants.  That inference is legitimate when it is argued case by case, and neither side can forbid it to the other.

Genealogy: why copies are not votes

Westcott and Hort's second great principle reads:  "all trustworthy restoration of corrupted texts is founded on the study of their history, that is, of the relations of descent or affinity which connect the several documents."¹  If fifty copies were made from one exemplar, they add up to one witness to that exemplar, plus whatever each copyist changed.  So Hort wrote that "a few documents are not, by reason of their mere paucity, appreciably less likely to be right than a multitude opposed to them."¹

Hort did not say numbers count for nothing.  In the next breath he granted "a theoretical presumption" that "a majority of extant documents is more likely to represent a majority of ancestral documents at each stage of transmission than vice versa."  He then judged it "too minute to weigh against the smallest tangible evidence of other kinds," because the number of copies made from each manuscript "varies irregularly from zero upwards."¹  Robinson builds his case on the first half of that paragraph, and quotes it.⁴  The disagreement between the schools is over how much that presumption weighs.  Part 6 prices it.

Robinson also reduces numbers by genealogy where it can be shown.  His example is "the ca. 124 MSS of Theophylact's commentary on John, which differ so little from one another that Theophylact's Johannine archetype readily can be reconstructed."⁴  One hundred and twenty-four copies, one witness.  On this point the schools agree.  They disagree about whether the Byzantine majority as a whole behaves like such a family, descended from a later common form, or like the normal product of centuries of copying from the autographs.

Part 3: Internal evidence: what the author wrote, and what scribes did

Two different questions

Internal evidence asks two questions about each rival reading.  The first is intrinsic probability:  which reading fits the author's vocabulary, grammar, style and argument?  The second is transcriptional probability:  if one reading was original, how did the other arise?  Transcriptional arguments can rest on observed scribal behaviour, which makes them less dependent on a modern reader's sense of what an apostle would say.  They are still probability judgments, and both kinds of argument can be turned around by a skilled opponent.

The master rule, accepted by every school, is to prefer the reading that best explains the rise of the others.  Robinson lists it first among his own principles and calls it "the guiding principle of the Nestle-Aland "local-genealogical" method."⁴  The schools part company over how to apply it.

What scribes did, and how it is measured

Scribal habits are studied mainly through singular readings, readings found in only one manuscript.  E. C. Colwell introduced the method on "the assumption that these readings are the creation of the scribe," and James Royse applied it to six early papyri in a 2008 study.¹⁴  The method has known limits, and Royse states them.  A reading singular today may have been inherited from a lost exemplar.  Hort already distinguished inherited singulars from "mere individualisms," and granted that telling them apart "may often be difficult or even impossible."¹⁴  Royse reduced the risk by checking each singular against far larger collations than Colwell used, so that "we can thus be even more confident that the readings we are left with were indeed created by the scribe."¹⁴  A second limit runs the other way.  A change two scribes made independently is not singular, so the method misses exactly the easy changes that recur.  A third is the sample:  all six papyri are Egyptian.  Royse grants that "if we want to know what kinds of changes were being made in Rome or Antioch, for example, we must turn elsewhere."¹⁴

The recurring kinds of change are these.

1. Sound-alike errors. In the Hellenistic period the Greek letters o and ō were pronounced almost alike, so ἔχομεν "we have" and ἔχωμεν "let us have" sounded the same.¹⁶ This matters at Romans 5:1.
2. Look-alike errors. Scribes wrote sacred names in short forms with a bar over them. "God" was ΘΣ and "Son" was ΥΣ, one letter apart.¹⁶ This matters at John 1:18. At Hebrews 2:9, "by grace" (ΧΑΡΙΤΙ) and "apart from" (ΧΩΡΙΣ) share their first, third and fourth letters, and Metzger's committee named "misreading χάριτι as χωρίς" as one possible route.¹⁶
3. Eye-skip. When two lines end alike, the eye can jump from the first to the second and drop the words between. Jumping back instead repeats words.
4. Harmonization. A scribe who knows a parallel passage, or who has just read a phrase two lines earlier, writes it again. Royse found that "the major influence on our scribes was the immediate context."¹⁴
5. Clarification and correction. Smoothing grammar, supplying a subject, fixing what looks like a mistake by the author. This matters at Mark 1:2.
6. Theological adjustment. Changing a phrase that could be misread by opponents. Ehrman's Orthodox Corruption of Scripture argues that scribes "occasionally altered the words of their sacred texts to make them more patently orthodox."¹³ How often this happened is disputed. Robinson's own fifth principle treats "religiously-motivated expansion and alteration" as secondary, so the schools agree that it happened.⁴

The two famous rules, and why each is only a tendency

"Prefer the harder reading."  Scribes tended to make the text easier, so the reading that looks harder is often the older one.  But nonsense is also hard, and nonsense is usually an error.  Robinson keeps the rule with that limit:  a scribe "would not deliberately produce nonsense."⁴  The rule weighs;  it does not decide.

"Prefer the shorter reading."  This rule assumed scribes added more than they cut.  Royse's measurements of the papyri, together with similar studies of other early manuscripts, found the opposite for the earliest period:  "the studies cited here provide substantial support to the thesis the scribes from the first several centuries tended to omit rather than to add."¹⁴  He warned against stretching this to "all scribes at all times."¹⁴  His proposed replacement reverses the default:  "In general the longer reading is to be preferred," except where the longer reading looks late on external grounds, or may come from harmonization, or from grammatical improvement.¹⁴  Robinson drops both defaults:  "Neither the shorter nor longer reading is to be preferred."⁴  On this point a leading Byzantine-priority scholar and the largest study of the papyri end up close together.

Is the reasoning circular?

Critics of the critical text say editors decide which manuscripts are good by how often they carry the "right" readings, and decide which readings are right by which manuscripts carry them.  Mink concedes the circle in so many words:  "witnesses are good because of their good variants, variants are good because of their good witnesses.  This circle cannot be avoided, but it has to be controlled."²  Three things limit it, and none removes it.

1. Scribal habits measured from singular readings depend less on the editor's choice of text than habits inferred from rejected readings, because no choice between rival readings is needed to find them. They still rest on the assumptions above.
2. Some accidental errors, such as eye-skip between like endings, have a direction that can be seen in the page layout and does not depend on theology.
3. The computer method of Part 4 checks whether thousands of individual decisions add up to a consistent family tree. Consistency is a real test, since a pattern of careless decisions tends to produce tangled families. It is not proof of accuracy, because a consistent set of decisions can still be consistently wrong.

A Byzantine-priority advocate faces a circle of his own, because the claim that the majority reflects normal transmission is also a premise used to judge readings.  Neither side escapes judgment.  The fair question is which judgments survive independent checks.

Part 4: The Coherence-Based Genealogical Method

The problem it was built for

Classical family-tree methods work when each copy has one parent.  New Testament scribes often had more than one exemplar, or corrected one copy against another, so a single manuscript can inherit readings from several lines.  This is called contamination or mixture, and it breaks simple trees.  Gerd Mink of the Institute for New Testament Textual Research in Münster designed the Coherence-Based Genealogical Method (CBGM) to cope with it.  Peter Gurry's summary of the difference:  "Where common error methods deduce ancestry, the CBGM aggregates it.  The fundamental principle of the CBGM is that the relationship of witnesses can be derived from the relationships of their variants."¹²

How it works

1. Collate. Record what every selected witness reads at every place of variation. For the Catholic Letters that was 123 witnesses and 3,046 places.¹²
2. Draw a local stemma at each place. The editors decide, place by place, which reading most likely gave rise to which. This is ordinary textual criticism, and early translations and church fathers can inform it, although they do not enter the later computer steps.³ Where the editors cannot decide, the local stemma leaves the source of a reading open, marked with a question mark.³ The method's users stress that "the existence of agreements between texts of manuscripts is primarily a matter of observation ... whereas the direction of disagreements is a matter of editorial judgment."³
3. Measure agreement. For every pair of witnesses, count the places where both are extant and how often they agree. This "pre-genealogical coherence" ignores direction. Vaticanus and Sinaiticus agree at 87.1 percent.¹²
4. Add up the directions. Wherever two witnesses differ, the local stemma may say whose reading came first. Across the Catholic Letters, the editors judged Vaticanus's reading prior to Sinaiticus's at 250 places and the reverse at 89, with 36 unclear and 15 not directly related.¹² Because Vaticanus has more prior readings, its text counts as a potential ancestor of Sinaiticus's text.³ Several texts can qualify as potential ancestors of the same witness. The editors' own reconstructed text, labelled A, takes part in this comparison as if it were a witness.
5. Draw the textual flow for each reading. At a given place, each witness is linked to its closest potential ancestor that has the same reading, searching only among its few closest potential ancestors. The size of that search is the connectivity: at a connectivity of ten, the computer "will only display an ancestor with a different reading if it cannot find an ancestor among the ten most similar ancestors for that witness."¹² If every witness to a reading finds a close ancestor with the same reading, the reading is coherent and may well have arisen once. If several witnesses can only be linked to ancestors with a different reading, the reading may have arisen more than once, or reached those witnesses by mixture. Wasserman and Gurry stress that the connectivity "is always determined in concert with our text-critical knowledge of the variant in question."³
6. Build substemmata and a global stemma. For each witness the method looks for the smallest set of potential ancestors that explains its text. This step was not finished for the Catholic Letters or Acts when Wasserman and Gurry wrote in 2017, and "a complete global stemma has yet to be produced."³³

The whole cycle is repeated.  Decisions that produce incoherent families are re-examined.  Mink describes the aim as a method "which can provide an overall view of the consequences of all the decisions we take."²

What the "ancestors" are

The CBGM relates texts, not physical books.  Wasserman and Gurry put it directly:  "the CBGM is not suggesting that Codex Sinaiticus (01) was directly copied from Codex Vaticanus (03).  The relationship is a more abstract one.  It is not a relationship between manuscripts but between the texts conveyed in them."³  A tenth-century minuscule can be a potential ancestor of a fourth-century codex, if its text preserves the earlier state more often.  Gurry adds that the flow diagrams "are not stemmas;  they are designed only for studying genealogical coherence, not the text's overall historical development."¹²

A concrete case: 1 Peter 4:16

Peter tells suffering Christians to glorify God either "in this matter" (ἐν τῷ μέρει τούτῳ, reading a) or "in this name" (ἐν τῷ ὀνόματι τούτῳ, reading b).  Reading b is in the early papyrus P72 and in Sinaiticus, Alexandrinus and Vaticanus.¹¹  Gurry notes that most editions before the new critical edition preferred b "because it is attested in so many early and important manuscripts and versions (P72, 01, 02, 03, 044, 33, 614, Latin, Coptic, Syriac)."¹²  Westcott and Hort printed it in 1881, and so does the SBL Greek New Testament of 2010.¹⁷  Reading a is the Byzantine reading, the Textus Receptus reading, and the King James Version's "on this behalf."¹⁸  The 2012 Nestle-Aland edition now prints a.¹⁹

The editors' reason is transcriptional.  Reading a "appears less natural and might easily have been changed by scribes to reading b, quite possibly under the influence of the phrase 'in the name of Christ' ... two sentences earlier."¹²  The flow data were then used as a check.  At a connectivity of ten, Gurry reports, eight witnesses to b "do not have close potential ancestors that share reading b with them":  03, 5, 1175, 876, 69, 88, 2718 and 218.  Reading a, by contrast, finds a same-reading ancestor for every witness.  He concludes that b "may have developed from reading a multiple times," and calls the case an example of how "the CBGM can confirm an editor's judgment."¹²¹²

This article reran the query on the Münster institute's public Genealogical Queries tool, which holds the 2013 Catholic Letters data.²⁰  It reproduces Gurry's eight witnesses at a connectivity of ten.  It also shows how the result depends on the setting.

Two further facts matter.  First, in this database Vaticanus has only one potential ancestor anywhere in the Catholic Letters:  the editors' reconstructed text A, which it matches at 96.0 percent of 2,971 places.²⁰  Once A is assigned "matter," Vaticanus must link to an ancestor with that reading.  This particular link adds no independent confirmation of the local choice.  The rest of the comparison still uses evidence from other places.  Second, the minuscules' lack of close "name" ancestors is real evidence about their relative similarity.  If the search is widened to fifty ranks, every minuscule finds a same-reading ancestor.  The rank limit expresses how close a textual relationship should be to support common descent at this particular place.  It is not a measured copying-error rate, and a rank of fifty does not mean fifty copying steps.²⁰¹²

The decision rests on a transcriptional judgment:  "matter" is the less natural phrase, and "the name of Christ" stands two verses earlier.  The flow data give conditional support:  under the editors' chosen rank limit, some "name" witnesses have closer potential ancestors reading "matter" than reading "name."  That fits repeated changes toward the easier phrase.  Broadening the search weakens this pattern, but finding a remote same-reading ancestor does not establish that it supplied the reading.  The data do not prove repeated origin.  A single early line carrying "name," shared by P72 and the great codices and mixed into later texts, remains a live alternative.

What the method cannot do

The method's own advocates list its limits.

1. It "is not a magic wand that will solve all our text-critical problems."³
2. It "is not making the decision for us; rather, it is providing a new type of evidence in the form of coherence to help us decide."³
3. In its application up to 2017, in the Catholic Letters and Acts, it "incorporates about one-third of our extant Greek manuscripts." Versions and fathers can inform the local stemmata, "but after that point we cannot see them in the textual flow diagrams or the global stemma." The global stemma "is more like a watercolor painting of a great national park than a topographical map."³
4. It inherits the editors' local judgments. If the editors misjudge many places in the same direction, the family tree will reflect the misjudgment consistently, and the 1 Peter 4:16 case shows how part of a flow result can simply restate a decision.

What it changed, and where the editions stand now

Applied to the Catholic Letters, the method "resulted in 34 changes to the main text" of the Nestle-Aland and United Bible Societies editions, published in 2012 and 2014.¹²  The editors also abandoned the old grouping of manuscripts into "text-types," except for the Byzantine text.¹²  Gurry reports that "almost one-third of the changes in the new NA/UBS text have the support of the Byzantine witnesses as over against readings found in witnesses from the third to fifth centuries."¹²  Wasserman and Gurry add that "ten of twelve changes between their first use of the CBGM on the Catholic Letters and their second use are in favor of the Byzantine text," and that in Acts, of fifty-two changes, "thirty-six cases ... were made in conformity with the Byzantine text and in only two cases against the Byzantine text."³

The method also counts the editors' own doubt.  Where they cannot choose between two readings, the major edition splits its main line, and the Nestle-Aland and United Bible Societies editions mark the place with a diamond.  At these places, Wasserman and Gurry explain, the editors "formally refrain from any rating."  They count 43 diamonds in the Catholic Letters, where the earlier editions had 32 bracketed places, and 155 in Acts, where they had 78.  Their own reading of the change is "a slightly greater uncertainty about the earliest text."³  Set against Gurry's 3,046 places of variation in the Catholic Letters, 43 undecided places is about one in seventy.¹²  A diamond marks only the places where the editors decline to choose at all.  Elsewhere they choose with less than certainty, so the count is a floor on doubt, not a full measure of it.  What the count does show is that the doubt has an address:  a reader can find every place where the editors could not decide, and both candidates are printed, one in the text and one in the apparatus.

The editions have moved on since.  The sixth edition of the United Bible Societies' Greek New Testament, published in 2025 with a corrected printing in 2026, takes in the major critical edition's volumes on Mark (2021), Acts (2017) and Revelation (2024).  "In these books, the present edition contains a new editorial text and fully revised apparatus," while "the text of the other biblical books remains unaltered."  Its preface also says that "Greater attention is paid to the Byzantine tradition, including the Textus Receptus."²¹  The publisher describes the matching 29th Nestle-Aland edition as being prepared, with no date given.²²  The Revelation volume's editors write that their team had the software and data for the CBGM, so that their text is "methodologically and materially analogous to the previous volumes."²³  In this article, John, Romans, Hebrews and the Petrine letters are therefore cited from Nestle-Aland 28, whose text in those books is unchanged.  Mark and Revelation are checked against the major edition's online apparatus.

Part 5: Seven places where the copies disagree

These cases were chosen because each tests a popular slogan, and several of those slogans are ones that serious advocates of every school already reject.  So the cases show how decisions are made and where they are weak.  They are not a random sample, and they cannot score the schools across the whole New Testament.

Greek readings were checked in the Münster institute's transcriptions, in the major critical edition's online apparatus for Mark and Revelation, and in the printed editions.  A transcription is a careful reading by trained staff, not an infallible image check, and where no transcription exists for a manuscript the article says so and does not guess its reading.  No complete apparatus was read for John, Romans, Hebrews or 1 and 2 Peter, so no totals of manuscripts are given for those books.

The last column matters for fairness.  Byzantine priority and the critical text agree at four of the seven places.  Revelation 22:19 divides the Textus Receptus from both of them, so it tests the Textus Receptus, not majority priority.

Mark 1:2: "Isaiah" or "the prophets"

The readings.  Most modern Bibles say "As it is written in Isaiah the prophet."  The King James Version says "As it is written in the prophets."²⁴  The quotation that follows is composite.  As Bruce Metzger noted, its first part comes from Malachi 3:1 and only the second from Isaiah 40:3.¹⁶

The witnesses.  The major critical edition of Mark lists "Isaiah the prophet," with or without the article, in Sinaiticus (01), Vaticanus (03), Bezae (05), Θ (038), the minuscule family led by 1, 33 and others, and in the Greek of Irenaeus and in Origen and other Greek fathers.  It lists "the prophets" in W (032), the minuscule family 13, many later manuscripts, the Byzantine tradition, the Latin translation of Irenaeus, and some versions.¹⁰  The Münster transcription of Alexandrinus (02) also reads "the prophets."¹¹  A few minuscules read "in the book of the words of Isaiah the prophet," which copies Luke 3:4.¹⁰  Robinson-Pierpont and the Textus Receptus print "the prophets."⁴

The case for "the prophets."  It is accurate, since two prophets are quoted.  It has the Byzantine tradition and two fifth-century codices.  And there is a plausible way for "Isaiah" to arise from it:  Matthew 3:3 and Luke 3:4 name Isaiah when they quote Isaiah 40:3, and scribes often brought one Gospel into line with another.  The Luke-shaped reading in a few minuscules shows scribes harmonizing at this very spot.

The case for "Isaiah."  There is an equally plausible way for "the prophets" to arise from "Isaiah":  a scribe who noticed that the first lines come from Malachi would want to fix the apparent mistake.  Metzger:  "It is easy to see, therefore, why copyists would have altered the words 'in Isaiah the prophet' ... to the more comprehensive introductory formula 'in the prophets.'"¹⁶  So both readings have an easy scribal motive, and either change could have been made once and spread, or made more than once.  The transcriptional arguments roughly cancel.  The difference lies in the external evidence.  "Isaiah" is the reading of the oldest Greek copies, of witnesses usually placed in separate groups, and of the Greek text of Irenaeus, a second-century writer.  "The prophets" first appears in Greek manuscripts of the fifth century, although Irenaeus's Latin translation shows the translator or a later copyist knew it.  Calling the "Isaiah" witnesses independent assumes the older groupings hold, and the editors of the critical edition have themselves dropped text-types except the Byzantine.  Their agreement still spans Egypt, the Western tradition and an early Greek father.

Decision.  Nestle-Aland, Westcott-Hort, the SBL edition and the major edition of Mark all print "Isaiah."¹⁷  This article's confidence is fairly high but not certain.  It rests on the age and spread of the "Isaiah" witnesses, not on a rule that the harder reading always wins.  No primary argument by a Byzantine-priority editor for "the prophets" at this verse was read for this article, which is a gap.

John 1:18: "God" or "Son"

The readings.  The verse ends the prologue:  "No one has ever seen God;  the only-begotten ... who is in the bosom of the Father, he has made him known."  The gap is filled by "God" (θεός) or "Son" (υἱός).  In the manuscripts these were abbreviated ΘΣ and ΥΣ, one letter apart.

The witnesses.  "God" is in the two oldest witnesses, the papyri P66 and P75, and in the first hands of Sinaiticus, Vaticanus and Ephraemi (04).  P75 has the article, "the only-begotten God";  P66, Sinaiticus and Vaticanus lack it.  A later corrector of Ephraemi changed "God" to "Son."  "Son" is in Alexandrinus and in the later majuscules Γ, Δ and Θ.¹¹  P66 and P75 are dated 200 to 225 in the Münster register, and P75 possibly to the early fourth century on Nongbri's argument.¹⁵¹¹  Bart Ehrman reports "Son" in "the Greek, Latin, and Syriac traditions" and in Irenaeus, Clement and Tertullian.¹³  This article did not check those patristic passages, so they are cited as Ehrman's report.

The case for "God."  It is the reading of the oldest Greek manuscripts.  The 1971 committee judged "Son," which "undoubtedly is easier," to be "the result of scribal assimilation" to John 3:16 and 18 and 1 John 4:9, where "only-begotten Son" occurs.¹⁶  On that view a scribe who knew John's usual phrase would write it here, and the change could happen in many places independently.  The committee also thought the anarthrous "God," without the article, "appears to be more primitive."  Some commentators read the phrase as three titles in a row:  "the only one, God, who is in the bosom of the Father."¹⁶  That reading removes the apparent clash of "the unique God" beside the Father.

The case for "Son."  Ehrman argued in 1993 that "the majority of manuscripts are right."¹³  Externally, he says "God" is found "almost exclusively in the Alexandrian tradition," while "Son" is found nearly everywhere else:  "both readings are ancient; one is fairly localized, the other is almost ubiquitous."  He adds that this "does not demonstrate" that "Son" is original.¹³  Internally, he argues that "Son" "coincides perfectly well" with how John uses the word elsewhere, and that "the unique God" makes poor sense in John:  "Jesus can be the unique God only if there is no other God; but for the Fourth Gospel, the Father is God as well."¹³  He grants that assimilation to John's usual phrase "is certainly a possibility," but thinks a change toward "God" fits an orthodox scribe's interest in Christ's deity.  He calls the three-titles reading "entirely implausible."¹³¹³  Within the 1971 committee, Allen Wikgren also dissented:  "It is doubtful that the author would have written μονογενὴς θεός, which may be a primitive, transcriptional error in the Alexandrian tradition (ΥϹ/ΘϹ). At least a D decision would be preferable."¹⁶

Decision.  Nestle-Aland, Westcott-Hort and the SBL edition print "God";  the Byzantine text and the Textus Receptus print "Son."¹⁷  In 1971 the committee rated its choice B, which its rating key defined as "some degree of doubt."¹⁶  That rating is historical:  the 2025 edition has "redefined" its ratings and changed some, and its new rating here was not read.²¹

Here the critical editions do follow the oldest attested Greek reading.  Ehrman's dissent shows a real disagreement among scholars, not a different choice by those editors.  The external evidence divides by region more than by age, since "God" is old in Egypt and "Son" is reported early in Gaul, Alexandria and North Africa.  Each internal argument has a mirror image.  Scribes could have conformed "God" to John's usual "Son," or heightened "Son" to "God."  A one-letter slip could run either way.  This article suspends judgment between the two readings.  The deity of Christ in John does not depend on this verse, since John 1:1 calls the Word God in every edition compared here.  What this verse says is still uncertain.

Romans 5:1: the critical text against its oldest codices

The readings.  "Since we have been justified by faith, we have peace with God" (ἔχομεν) or "let us have peace with God" (ἔχωμεν).  The two words differ by one vowel, o or ō, which sounded almost alike.

The witnesses.  The first hands of Sinaiticus and Vaticanus wrote the subjunctive, "let us have";  later correctors of both changed it to the indicative.  Alexandrinus, Ephraemi and minuscule 33 have the subjunctive.  The third-century parchment 0220 has the indicative, though its transcribers mark two of the key letters as uncertain.¹¹  Metzger lists minuscule 1739 for the indicative.¹⁶  The Münster transcription instead reads its first hand as probably subjunctive, with the omega marked uncertain, corrected to the indicative.¹¹  The images of that page are not public, so this article cannot settle which is right and does not count 1739.  The Byzantine text and the Textus Receptus have the indicative.⁴

The case for each.  The subjunctive has, in Metzger's words, "far better external support."  The committee printed the indicative anyway, because "Paul is not exhorting but stating facts ... only the indicative is consonant with the apostle's argument," and because the two words sounded alike, so "when Paul dictated ἔχομεν, Tertius, his amanuensis (16.22), may have written down ἔχωμεν."¹⁶  Westcott and Hort followed the oldest manuscripts and printed the subjunctive.¹⁷

A point about targets.  The Tertius suggestion has a consequence the committee did not spell out.  If the secretary wrote ἔχωμεν and Paul sent the letter, then the autograph itself read ἔχωμεν, and the indicative is what Paul meant rather than what the first copy said.  The choice then aims at the author's intended wording, which is a fifth possible target beside the four in Part 1.  On the other route, a later copyist's slip, the indicative would be the autograph's reading too.  The two explanations lead to the same printed word for different reasons.

Decision.  The 2012 Nestle-Aland text prints "we have."¹⁹  The committee's 1971 rating was C, "a considerable degree of doubt."¹⁶  Here the critical text agrees with the Byzantine text against the first hands of its oldest codices, on internal grounds.  The case is one of low confidence.

Hebrews 2:9: "by the grace of God" or "apart from God"

The readings.  Jesus tasted death for everyone "by the grace of God" (χάριτι θεοῦ) or "apart from God" (χωρὶς θεοῦ).

The witnesses.  "By the grace of God" is in P46, dated to the early third century, and in Sinaiticus, Alexandrinus, Vaticanus, Ephraemi, 33 and 81.¹¹¹¹  It is the Byzantine and Textus Receptus reading.⁴  Metzger lists "apart from God" in 0121b (now numbered 0243), in the first hand of 1739 and in a corrector of 424, in some Latin and Syriac copies, and in "a rather large number of Fathers, both Eastern and Western."¹⁶  The Münster register dates 0243 and 1739 to the tenth century and 424 to the eleventh.¹¹  No Münster transcription of those three manuscripts at this verse was available, so their readings rest on Metzger's list.

The key early witness is the third-century scholar Origen.  In his Commentary on John he writes, in Allan Menzies's translation:  "For without God He tasted death for every one. In some copies of the Epistle to the Hebrews the words are by the grace of God."²⁵²⁶  So Origen knew both readings, quoted "apart from God" as his text here, and reported "by grace" in "some copies."  Ehrman goes further and calls "apart" the reading of "the majority of manuscripts" of Origen's day.¹³  Origen's words do not count copies, and one citation chosen for one argument does not measure what was common in his library or his region.  Erwin Preuschen, the editor of Origen's Greek text, also notes that Origen elsewhere quotes the verse with "grace" first and "apart" second.²⁶  What Origen does establish is that "apart from God" was known and used in the third century, long before any surviving Greek manuscript that has it.

The case for "apart from God."  It is the harder reading by far.  Ehrman argues that a scribe would not turn a common phrase into a rare one:  "Is he likely to have created a phrase that never occurs elsewhere in the New Testament (χωρὶς θεοῦ) or one that occurs over twenty times (χάριτι θεοῦ)?"¹³  A scribe might also have found "apart from God" theologically awkward and replaced it.

The case for "by the grace of God."  It has the earliest manuscript, P46, and nearly all the rest.  Metzger's committee offered two routes to "apart":  "a scribal lapse, misreading χάριτι as χωρίς, or, more probably, ... a marginal gloss (suggested by 1 Cor 15.27)," a note that "everything" in verse 8 does not include God, later copied into verse 9 by mistake.¹⁶  Ehrman replies that the gloss theory is "altogether too clever, and requires too many dubious steps to work."¹³  On the simpler misreading route he grants "This is a possibility," but argues that a slip of that kind "would more likely have occurred in the reverse direction," from the rare word to the common one.¹³  Each side therefore has a plausible mechanism, and neither mechanism is observed.

Decision.  Nestle-Aland and the Byzantine text print "by the grace of God."  The SBL Greek New Testament, edited by Michael Holmes, prints "apart from God."¹⁷  This is an open case.  It shows that a reading now rare among Greek manuscripts was known to a leading third-century scholar, so present-day proportions need not match early ones.  It does not show what the third-century proportions were.

1 Peter 4:16: a change toward the Byzantine reading

This case was set out in Part 4.  The current critical text agrees with the Byzantine text and the Textus Receptus.¹⁸¹⁹  The editors' reason is that "in this matter" is the less natural phrase and "in the name of Christ" stands two verses earlier.  The flow data give conditional support under their chosen rank limit.  The case shows that the critical editors will print a Byzantine reading against the oldest manuscripts when they judge the internal evidence strong enough.  Confidence is moderate.

2 Peter 3:10: a reading with no Greek manuscript

The readings.  On the day of the Lord, "the earth and the works in it" will be "found," "burned up," "made to disappear," "found dissolved," or "not found."

The witnesses.  Sinaiticus and Vaticanus have "will be found."  P72 has "will be found dissolved."  Alexandrinus and minuscule 33 have "will be burned up," as do the Byzantine text and the Textus Receptus.  Ephraemi has "will disappear."⁴¹¹  Metzger reported that "the Sahidic version and one manuscript of the Harclean Syriac version insert the negative," producing "will not be found."¹⁶

The case for each.  "Will be found" is the oldest Greek reading, and each of the others looks like an attempt to make sense of it.  But it seems to say the opposite of what the passage needs.  Metzger's committee printed it in 1971 with the lowest rating, D, while judging that none of the readings "seems to be original," and noting that "various conjectural emendations have been proposed."¹⁶  "Burned up" makes clear sense and has the Byzantine majority, but it looks like the kind of clarification scribes made.

Decision.  Westcott-Hort, the SBL edition and the 27th Nestle-Aland edition print "will be found."¹⁷  The 2012 edition prints "will not be found."¹⁹  Wasserman and Gurry note that this reading "is only attested in a few Coptic and Syriac manuscripts (i.e., only in translations); there is no Greek manuscript support at all, which is quite exceptional."³  They report Mink's judgment that it "is demanded by the logic of the passage, making it an 'almost unavoidable conjecture,'" and the editors' local stemma derives every other reading from it, directly or through another reading.³

Robinson's principle runs the other way:  "The quantity of preserved evidence for the text of the NT precludes conjectural emendation."  His reason is that a conjecture requires "an unparalleled transmissional catastrophe in which all known witnesses ... failed to preserve the original text," and he counts manuscripts, translations and church fathers alike as witnesses.  He judges the chance that a conjecture restores the original "virtually nil."⁴  By his definition 2 Peter 3:10 is a narrower case, because two translations keep the reading.  The 2012 decision assumes that every surviving Greek copy lost the right reading here and only translations kept it.  Neither judgment can be checked against an autograph.  The editors' reasons are printed and can be argued with.  That is to their credit, but it does not by itself show their judgment is right, and a skeptic of either school can ask whether a preference for a sensible text drove it.  The case deserves low confidence, and the 1971 committee gave its own choice the lowest rating it had.

The editions' own record supports Robinson's description that conjecture "does not gain a serious foothold in contemporary praxis."⁴  Wasserman and Gurry note that 2 Peter 3:10 "is the only conjecture in the ECM of the Catholic Letters."³  Peter Gurry's list, compiled with Jan Krans of the Amsterdam database of conjectures, names three printed in the current editions:  Acts 16:12 in Nestle-Aland 27 and 28, 2 Peter 3:10 in Nestle-Aland 28, and Acts 13:33 in the major edition of Acts.  He concludes that the Nestle editions "have grown more reticent to print conjectures since 1927 not less."²⁷  At Acts 13:33 the major edition's main line reads "to us, the children" (ἡμῖν) in a form its apparatus lists for no Greek manuscript.  At Acts 16:12 it went back to the manuscripts' reading, "a leading city of the district," found in P74, Sinaiticus, Alexandrinus and Ephraemi.¹⁰¹⁰  The 1971 committee had printed a conjecture there by majority vote, over a signed dissent by Kurt Aland and Bruce Metzger that "it appears ill-advised to abandon the testimony" of those manuscripts.¹⁶  This count shows how rarely editors who are free to conjecture conclude that the whole Greek tradition lost a reading.  It cannot show how often a loss happened without leaving an awkward text behind, since only an awkward text prompts a conjecture.

Revelation 22:19: where the Textus Receptus leaves the Greek tradition

The readings.  The King James Version says God shall "take away his part out of the book of life."  Modern versions say "tree of life."²⁴

The witnesses.  The major critical edition of Revelation, published in 2024 and updated online, cites "tree" (ξύλου) in Sinaiticus, Alexandrinus, the tenth-century majuscule 046 and about seventy minuscules, including 141.  It cites "book" (βιβλίου) in minuscule 61 and in a correction to minuscule 2067, and "the part of the book" in the first hand of 61.¹⁰  The Münster register dates 61 to the sixteenth century and 2067 to the fifteenth;  the correction in 2067 is not dated.¹¹  Several early witnesses, including P47 and Ephraemi, are not extant here.  The Byzantine text and the critical text print "tree."⁴¹⁹  The Textus Receptus reads "book."¹⁸  The Latin tradition has "book":  one strand of the Old Latin reads libro, and Hills cites the Vulgate.⁹²⁸

How the Textus Receptus got its reading.  Erasmus had a single Greek manuscript of Revelation for his first edition in 1516, and "this manuscript lacked the final leaf, which contained the last six verses of the book."  He "depended on the Latin Vulgate and translated the missing verses into Greek."²⁹  Martin Karrer, an editor of the major edition, writes in German that Erasmus's guesses where his manuscript had gaps carried a high risk of error, because he had no earlier copy and no parallel manuscript.  His example, in translation:  "Erasmus wrongly wrote βίβλου instead of ξύλου τῆς ζωῆς in 22:19."³⁰

The best defence of "book."  Edward Hills's 1956 edition accepted that Erasmus "manufactured" some Greek and called such readings "minor blemishes."³¹  His 1984 edition argues more strongly.  He grants that "According to Hoskier, all the Greek manuscripts, except possibly one or two, read, tree of life."  He then defends "book" from "the Latin Vulgate (including the very old Vulgate manuscript F), the Bohairic verson, Ambrose (d. 397), and the commentaries of Primasius (6th century) and Haymo (9th century)."  And he questions the back-translation itself:  "Hoskier seems to doubt that Erasmus did this, suggesting that he may have followed Codex 141."⁹

Testing the defence.  The Latin support is real.  The proposed Greek source fails the check:  the major edition cites manuscript 141, a fourteenth-century copy, for "tree."¹⁰¹¹  The two Greek witnesses to "book" are a sixteenth-century manuscript and an undated correction.  Neither has been established here as independent of Erasmus's printed text of 1516.  The same edition's editors set aside other late copies "based in part or entirely on the printed Erasmian text."²⁸  That does not establish that 61 or the corrector of 2067 copied print at this verse.  In particular, the correction's unknown date prevents ruling out a pre-1516 reading.  The narrower finding is that neither witness securely establishes one.  An early Greek source for "book" remains possible, but 141 does not supply it.

Decision.  High confidence for "tree," on the agreement of every extant early Greek witness and the Byzantine tradition.  This case divides the Textus Receptus position from the Byzantine-priority position.  A Byzantine-priority editor prints "tree."  Hills's defence of "book" must rely mainly on versions and fathers, rather than on the Greek majority.

Part 6: The schools in their own words, tested

Reasoned eclecticism

This is the method behind the Nestle-Aland and United Bible Societies editions, the SBL edition and most modern translations.  It weighs external and internal evidence place by place, guided by what is known of the history of the text, and since 2012 the main edition has checked its decisions with the genealogical method of Part 4.  In the cases above it printed a reading against its oldest codices at Romans 5:1, moved to the Byzantine reading at 1 Peter 4:16, and printed a reading with no Greek manuscript at 2 Peter 3:10.  That range is consistent with weighing evidence rather than following one rule.  Its weakness is the one Mink admitted:  judgments about readings and judgments about manuscripts feed each other.²  Its editors also disagree among themselves at hard places, as Hebrews 2:9 and 2 Peter 3:10 show.

Byzantine priority

Maurice Robinson, co-editor of the Robinson-Pierpont edition, calls his method "reasoned transmissionalism":  it "evaluates internal and external evidence in the light of transmissional probabilities," and its "overriding principle is that textual criticism without a history of transmission is impossible."⁴  He rejects the caricature of counting.  A purely numerical method, he notes, "would offer no solution in the many places where the Byzantine Textform is divided among its mass of witnesses."⁴  His preface adds that "Quantity alone cannot be determinative when evaluating variant units."⁴  He reduces numbers by genealogy where it can be shown, weighs manuscripts, rejects "an exclusive following of the oldest MSS," and rejects conjectural emendation.⁴⁴⁴⁴

His central claim is about history.  The Byzantine text "predominated among the Greek-speaking world from at least the fourth century until the invention of printing," and any theory must explain why.⁴  He offers the parallel of Homer, whose medieval "vulgate" text seems to have existed before the Alexandrian scholars, and he suggests that the shorter Alexandrian New Testament text "may well reflect recensional activity."⁴  On this view the great Egyptian manuscripts preserve a local, edited branch, and the majority preserves the main trunk.

What supports it.  The critical editors' own reappraisal:  they found Byzantine witnesses "surprisingly similar to their own reconstructed text," now call the Byzantine text "an important witness to the early text," and made most of their later changes toward it.³  The newest United Bible Societies edition pays "greater attention" to the Byzantine tradition.²¹  None of this shows the Byzantine text is the autograph text, and none of it shows it is not.  It shows that its readings are now judged one by one rather than discounted as a group.

What counts against it.  By Robinson's own account the Byzantine text first appears in Greek manuscripts of the fifth century, Alexandrinus and W, while a text very close to Vaticanus's is already in P75.⁴  The force of that point depends on P75's date:  early third century on the Münster register, and possibly early fourth on Nongbri's argument, which would bring P75 close to Vaticanus itself.¹⁵¹¹  Robinson's reply is that the change from papyrus to parchment in the fourth century, and from capital to minuscule script in the ninth, led scribes to recopy and discard older books, so that "many early minuscules are likely only two or three generations removed from papyrus ancestors of the fourth century or before."⁴  That is possible.  It is an inference about lost books, the same kind of inference the critical text is accused of making, and the surviving evidence does not settle it either way.  At Mark 1:2 the Byzantine reading has later and narrower early support than its rival.  The probability argument for the majority is examined below.

Burgon and the "traditional text"

John William Burgon, Dean of Chichester, defended the Byzantine-type "Traditional Text" against Westcott and Hort.  His seven "Notes of Truth" are, in order:  "Antiquity, or Primitiveness";  "Consent of Witnesses, or Number";  "Variety of Evidence, or Catholicity";  "Respectability of Witnesses, or Weight";  "Continuity, or Unbroken Tradition";  "Evidence of the Entire Passage, or Context";  and internal considerations.³²  He thought the slogan that witnesses "are to be weighed— not counted" embodies "much fundamental fallacy," and held that "Number then constitutes Weight ... Not of course absolutely, as being the sole Test, but caeteris paribus, and in its own place and proportion."³²  This is a weighing method, not a vote.  Where Burgon and the modern editors differ most is in what they think of Sinaiticus and Vaticanus.

The Majority Text of Hodges and Farstad

Zane Hodges and Arthur Farstad's edition of 1982, revised in 1985, is a separate project from Robinson's.  They describe it as "a text that employs the available evidence of the whole range of surviving manuscripts rather than relying chiefly on the evidence of a few."³³  They hold that genealogy is "the most viable and promising option" where the evidence divides, but that it "is presently possible only in the Pericope Adulterae and in Revelation."³³  In Revelation, they report, "the Majority Text agrees with the Critical Text more than twice as often as it conforms to the Textus Receptus," which they put down partly to "the TR's being originally based on a very narrow manuscript foundation."³³  So the majority-text editors themselves treat the Textus Receptus as a different text from theirs.

Pricing the majority argument

The best-known mathematical defence of the majority is an appendix by Zane Hodges and his brother David Hodges, a statistician, printed in Wilbur Pickering's The Identity of the New Testament Text.  Zane Hodges states the intuition:  "Under normal circumstances the older a text is than its rivals, the greater are its chances to survive in a plurality or a majority of the texts extant at any subsequent period. But the oldest text of all is the autograph."⁶  He adds a condition, "barring some radical dislocation in the history of transmission."⁶

The model, in its authors' words.  David Hodges assumes that "good manuscripts and bad manuscripts will be copied an equal number of times, and that the probability of introducing a bad reading into a copy made from a good manuscript is equal to the probability of reinserting a good reading into a copy made from a bad manuscript."⁶  The appendix is candid about the limits.  It says that "if one type of manuscript is copied more frequently than the other, the type of manuscript copied most frequently will perpetuate its reading more frequently";  that if the two probabilities differ, "the discussion does not apply";  and that scribes consulted the copies nearest to hand, so "there would be branches of texts which would be corrupt."⁶  Zane Hodges calls the set-up an "'idealized' situation which does not represent what actually took place."⁶  The calculation below treats each copying event's change as independent of other copying events, conditional on the exemplar's reading.  Copies sharing an ancestor still inherit its reading;  they are not independent votes.

Its two claims.  The first is about the average.  With a one-in-three error rate and three copies per manuscript, "the original majority for the correct reading is more likely to be retained than lost. But the majority in the fifth generation is a slender 41:40."⁶  That figure is exact.  The expected share of good copies moves toward one half every generation and settles there, and the appendix's 41:40 is the fifth step.⁷  The second claim is about a particular split, and it is the one the argument needs:  "Given about 500 manuscripts of which 75% show one reading and 25% another, given a one-third probability of introducing an error, given the same probability of correcting an error, and given that each manuscript is copied twice, the probability that the majority reading originated from an error is less than one in ten."⁶  The appendix prints no calculation for it.

The second claim, computed.  This article computed the full probability distribution of the model, not just its average.⁷  Three ways of counting about 500 copies were tried:  the whole last generation, every copy of every generation, and 500 survivors drawn at random from a larger last generation.  Two and three copies per manuscript were both tried, producing groups of 500, 510, 512 or 729 copies.  The table uses a band around the claimed split:  one reading at 75 to 85 percent and the other at 15 to 25 percent.  Its ranges span those counting choices.  Exact 75:25 splits, rounded to whole copies, give very similar conditional results and the same verdict on the claim.

So the appendix's conclusion holds for careful copying and fails at the one-in-three rate it printed.  At that rate a lopsided split is extremely rare.  It gives almost no preference between the readings when only the last generation survives, and a modest preference for the majority when copies from earlier generations also survive.  Neither result reaches the claimed probability above 90 percent for the majority.  At low error rates a lopsided split strongly favours the majority.  The appendix's sentence "If the probability of introducing an error is less than one-third, the probability ... is even less" points the right way.⁶

What the cumulative argument needs.  Hodges grants that "an error might easily be copied more often than the original reading in any particular instance," but asks whether in "100 readings where the manuscripts divide 80 percent to 20 percent" we are to suppose "this reversal of probabilities has occurred" again and again.⁶  Counting those outcomes as independent reversals requires more than the single-reading model.  Errors at different places can occur independently on a shared family tree;  common ancestry alone does not make them correlated.  But if one copy carrying many departures from the autograph was copied or preserved disproportionately, its departures could become common together.  That would be one shared success in copying or survival, rather than a hundred independent ones.  No rate for such an event is established here.

A separate illustration shows why grouping known relatives matters.  Suppose the reading of one early ancestor E and one later ancestor L has been securely recovered from their descendants, and E and L came down separate lines from the autograph.  Once those two ancestral readings are known, counting more descendants does not add votes for either ancestor.  Under the same symmetric error rule, the chance that E rather than L retained the original depends on the copying steps between each ancestor and the autograph.⁷  This is a changed scenario, not another result from the equal-copying tree above, and it does not show that the Byzantine witnesses form such a family.  Hodges's answer is historical.  He argues that no one has explained how a slow "process" could produce so uniform a majority, and that "an unguided process" achieving it "imposes impossible strains on our imagination."⁶  The critical editors' answer is also historical.  Neither answer is supplied by counting.

The low-error calculation supports a majority presumption under equal copying.  Applying it to the real tradition also requires a case about unequal copying, survival, mixture and the relation between different readings' histories.  That is the question of where the Byzantine text came from, and it is argued from the papyri, the versions, the fathers and the readings themselves.

Part 7: Preservation as belief, transmission as record

The theological claims, stated by their holders

The Textus Receptus and King James-only positions start from a doctrine, and they say so.  The Westminster Confession of 1646 states that the Old Testament in Hebrew and the New Testament in Greek, "being immediately inspired by God, and, by his singular care and providence, kept pure in all ages, are therefore authentical."⁸  The Confession names no printed edition.

Edward Hills built a textual method on that doctrine.  In 1956 his axioms began:  "(a) The purpose of the providential preservation of the New Testament is to preserve the infallibility of the inspired original text," and reached "(e) The text of the majority of the manuscripts is the providentially preserved and approved text."³¹  In his 1984 fourth edition he called his reasoning "the logic of faith":  "Because the Gospel is true, the Bible which contains this Gospel was infallibly inspired by the Holy Spirit. And because the Bible was infallibly inspired, it has been preserved by God's special providence. Moreover, this providential preservation was not done privately in secret holes and caves but publicly in the usage of God's Church. Hence the true New Testament text is found in the majority of the New Testament manuscripts. And this providential preservation did not cease with the invention of printing. Hence the formation of the Textus Receptus was God-guided."⁹

The Trinitarian Bible Society, which publishes Scrivener's 1894 Textus Receptus, grants that "There is no single Received Text" and that Erasmus "used a handful of manuscripts."  It argues that the editions' manuscripts "were all of the Byzantine text-type," so that the editions "are nearly consistent, not only with one another, but also with the vast majority of manuscripts."⁵⁵  D. A. Waite, a King James-only writer, goes further:  "The inspired, preserved Hebrew, Aramaic, and Greek Words underlying the King James Bible are perfect Words and are a perfect Bible."³⁴

What the manuscripts can and cannot test

A belief that God preserved his word, stated with no further detail, fits any manuscript record, because it does not say where or how the preserved text is to be found.  No collation can confirm or refute it, any more than it can confirm or refute providence in the rest of history.

Specific forms of the belief are different, because they say where to look.  Hills's form says providence worked "publicly in the usage of God's Church" and is seen in the majority of manuscripts.  That predicts that the true readings will be found in the Greek manuscripts the church used.  It also allows a check:  where the Textus Receptus departs from that majority, Hills's own premise counts against the printed text, unless a further claim about providence in printing is added, which Hills does add.  Waite's form says the words "underlying the King James Bible" are perfect.  That requires an answer to which Greek text that is, since the King James translators followed no single printed edition.  The rest of this part tests the factual claims that named defenders make.  It does not test the doctrine itself, and it does not treat faith as a reason to set aside a defender's historical arguments.

The factual claims, tested

1. Claim:  the Textus Receptus editions are nearly consistent with the majority of manuscripts (Trinitarian Bible Society).  Partly true, and a matter of degree.  Robinson's Byzantine edition counts "over 1800 instances" where the printed Textus Receptus differs from the Byzantine text, "generally due to the inclusion of weakly supported non-Byzantine readings."⁴  Across the whole New Testament that is close agreement, but not identity.  Hills himself granted non-Byzantine readings in the Textus Receptus and defended some of them, including the Comma Johanneum, which he called "possibly genuine," and the Textus Receptus's placement of the doxology in Romans:  "Erasmus seems to have been right in following the Latin Vulgate rather than the Byzantine text."³¹³¹  At those places a Textus Receptus defender is not defending the majority.

2. Claim:  the Textus Receptus is one fixed text.  The Trinitarian Bible Society does not claim this.  Its own statement is that "There is no single Received Text."⁵  The editions differ.  Robert Stephanus's edition of 1550 ends the angel's address at Revelation 16:5 with ὁ ὅσιος, "the holy one."³⁵  Scrivener's 1894 text, built to match the King James Version, has ὁ ἐσόμενος, "who will be," which is what the King James Version translates:  "which art, and wast, and shalt be."²⁴¹⁸  The major critical edition of Revelation records thirteen Greek forms of this phrase, every one with "holy," and lists "who will be" with no Greek witness at all.¹⁰  Scrivener made his reconstruction because "the Authorised Version was not a translation of any one Greek text then in existence, and no Greek text intended to reproduce in any way the original of the Authorised Version has ever been printed."  He took Beza's edition of 1598 as his base and listed "All variations from Beza's text of 1598, in number about 190."  He also noted places where "the Authorised Version corresponds but loosely with any form of the Greek original, while it exactly follows the Latin Vulgate."¹⁸¹⁸  The name "Textus Receptus" itself comes from a publisher's preface of 1633, which told the reader he had "the text now received by all, in which we give nothing changed or corrupted."  Metzger and Ehrman call it a "blurb."²⁹

3. Claim:  Erasmus's Greek at Revelation 22:19 may have had a Greek source (Hills, 1984, after Hoskier).  Not supported by the current evidence.  The manuscript proposed, 141, reads "tree."  The only Greek witnesses to "book" are a sixteenth-century manuscript and an undated correction.⁹¹⁰  Metzger and Ehrman also report that at Acts 9:6 Erasmus added words from the Latin "found in no Greek manuscript at this passage."²⁹  Robinson, for his part, rejects the search for a "providentially preserved" printed text as firmly as any critical editor:  "such an approach has nothing to do with actual text-critical theory or praxis."⁴

4. Claim:  the King James translators claimed a perfect English text.  Their own preface does not.  It argues that "the very meanest translation of the Bible in English ... containeth the word of God, nay, is the word of God," a claim that covers every faithful translation, not one.³⁶  They also defended printing alternatives in the margin:  "Some peradventure would have no variety of senses to be set in the margin, lest the authority of the Scriptures for deciding of controversies by that show of uncertainty, should somewhat be shaken. But we hold their judgment not to be sound in this point."³⁶  This tests a historical claim about the translators.  It does not test whether later believers may hold the translation in higher regard than its makers did.

Peter Ruckman and Gail Riplinger make stronger King James-only claims than Waite's.  Their books were not obtained for this article, so their arguments are not assessed here.

Doctrine and wording are different questions

A doctrine can be taught clearly in many passages while one passage's wording stays uncertain.  The deity of Christ does not depend on whether John 1:18 says "God" or "Son," because John 1:1 says "the Word was God" in every edition compared for this article.³⁷  The reverse also holds.  A doctrine being true elsewhere is not evidence that a disputed verse contains it.  Defenders of different texts sometimes argue from clarity of doctrine.  The Trinitarian Bible Society asks:  "Where else in Scripture is the Trinity so clearly delineated as in 1 John 5.7-8?"⁵  Hills wrote in 1956 that non-Byzantine readings "ought never to be preferred in places in which they ... detract from its doctrinal richness."³¹  Both choose readings partly by theological value, which is the charge such defenders bring against critical editors.  Whether a doctrine is true and whether a scribe wrote a word are separate questions, and each needs its own evidence.

Objections and responses

1. "Most manuscripts agree, so the majority reading should win."  Copies of one ancestor are one witness to that ancestor, a point Robinson accepts as well as Hort.⁴  The majority presumption is real but conditional.  Computed exactly, the Hodges model strongly favours the majority when copying is careful and equal, but gives much weaker support at the error rate its authors chose.  Preferential copying or survival of one ancestor can make many of its readings common together.  Whether that happened is a question of history, not of the count.

2. "Modern Bibles rest on two Egyptian manuscripts corrupted by heretics."  Burgon accused Sinaiticus and Vaticanus of a "sceptical character" and an alliance with "the Origenistic School."³²  The critical text in the cases above does not simply follow those two:  it departs from their first hands at Romans 5:1 and from both at 1 Peter 4:16 and 2 Peter 3:10.  Whether the early Egyptian text is an edited local branch is a fair question, and agreement percentages alone cannot settle it.  The charge of heretical corruption needs specific readings shown to be heretical changes, and the cases here do not supply them.

3. "Editors pick readings to suit their doctrine."  Motives cannot be read off a choice, and published reasons do not prove freedom from bias.  What can be done is to test the reasons.  At John 1:18 the critical text prints the reading with the higher Christology, and the case for it rests on the oldest Greek witnesses and on assimilation;  Ehrman's case against it rests on spread and Johannine usage.  At Hebrews 2:9 the critical text keeps "grace," and the case for "apart" is the harder-reading argument.  In each case the reasons can be checked against the evidence, and this article finds both open.  A doctrinal argument for a reading is equally suspect on either side, as Part 7 shows.

4. "Scribal tendencies are circular:  you decide what scribes did from the readings you have already rejected."  The main measurements use singular readings, which do not require choosing between rival readings, though they rest on their own assumptions.¹⁴  The CBGM tests whether thousands of separate decisions fit together.  Neither removes the circle.  As Mink says, it "cannot be avoided, but it has to be controlled."²  The fair test is whether a method's judgments survive checks that do not depend on them.

5. "There are no originals, so the original text is unknowable."  The editors themselves aim at the initial text and grant it may differ from the autograph.²  What can be measured is how stable the text is from the earliest evidence onward, and where the doubts are, and three measurements point the same way.  First, where one line of copying can be followed across time, it barely moved.  Gordon Fee found that the relationship between P75 and Vaticanus in John "is far closer than the relationship between any two other New Testament MSS of any kind in the first four Christian centuries."  In Luke 10 the two disagree twelve times, against eight and fourteen for the closest pairs in Family 1, a medieval family of copies.³⁸  The time this spans depends on P75's date:  a century or more on the Münster register, much less on Nongbri's argument.¹⁵  Second, the editors' doubt is counted and placed:  in the Catholic Letters they decline to choose at 43 places, against Gurry's 3,046 places where the copies differ (Part 4).³  Third, editors who are free to conjecture almost never conclude that every Greek copy lost the initial text:  once in the Catholic Letters, at 2 Peter 3:10, where translations keep the reading, and at a handful of places elsewhere.³²⁷  None of this reaches behind the earliest surviving copies.  A change made before them that left no trace would leave no awkward reading to prompt a conjecture and no split to mark with a diamond.  That gap is real, and no method can measure it.  The variant counts and the debate over the earliest period are in the sibling article on manuscripts.

6. "Critical editions keep changing and disagree with each other, so the text is unstable."  The changes are public, and most are small.  The genealogical method changed 34 places in the Catholic Letters and 52 in Acts.³  Gurry judges that "in most cases the changes are of minor significance for interpretation or translation, but in several cases the changes should not be ignored," and names Jude 5, where the text now says "Jesus" saved a people out of Egypt.¹²  The 2025 United Bible Societies edition has new texts for Mark, Acts and Revelation and left the other books unchanged.²¹  Leading editions disagree at hard places such as Hebrews 2:9 and 2 Peter 3:10.  The printed Textus Receptus also varies between its editions, as Revelation 16:5 shows, and its publishers say so.

7. "Printing a conjecture at 2 Peter 3:10 means scholars are writing Scripture."  The reading is not invented.  It is in the Sahidic Coptic version and a Syriac manuscript.¹⁶  But it has no Greek manuscript, and Robinson's objection that the volume of evidence should rule out such choices deserves more answer than the editors have given.⁴  It is the least certain kind of decision in the edition.

8. "The King James Version is the preserved word in English, so the Greek should be read through it."  The translators defended marginal alternatives and said the meanest translation was the word of God.³⁶³⁶  An English-first rule also has to say which Greek the English translates.  At Revelation 16:5 the ending translated "shalt be" has no Greek witness in the major edition.  At 22:19 "book" has only late or undated Greek support there;  its presence before Erasmus's edition has not been established.

9. "If this verse is uncertain, the doctrine it teaches is uncertain."  A doctrine taught in many secure passages does not depend on one insecure passage.  The reverse also holds:  a doctrine's truth elsewhere is no evidence for a particular wording here.  Part 7 gives examples.

10. "Byzantine priority is a fringe view and should not be called serious."  Its leading defender states a full method, accepts genealogy where it can be shown, and weighs internal evidence.⁴  The critical editors have themselves moved toward Byzantine readings and call the Byzantine text an important witness to the early text.³  In this article's cases the critical text and the Byzantine text agree at four of seven places.  Of the three where they differ, one is settled with fairly high confidence and two are open.

11. "Testing a providentially preserved text by manuscripts is unbelief."  The Westminster Confession names no printed edition.⁸  Hills tested readings against manuscripts himself, argued from Latin, Coptic and patristic evidence at Revelation 22:19, and appealed to a Greek manuscript.⁹  Once a defender makes such factual claims, checking them is taking the defender seriously.

Issues and unanswered questions

1. The first copies are not observed. Mink grants that changes between the autograph and the initial text "may not have left a single trace."² Nothing in this article can measure what happened before the earliest witnesses. The counts of undecided places and of printed conjectures measure only the losses and doubts that show. A clean loss would leave neither. The conjecture counts here come from the Catholic Letters volume and from Gurry's 2018 list; the Mark (2021) and Revelation (2024) volumes were not searched for conjectures.
2. The genealogical method covers part of the New Testament. Its published results here are for the Catholic Letters, Acts, Mark and Revelation. In the 2017 account it used about a third of the Greek manuscripts and carried no versions or fathers into its computer steps.³ The Gospels other than Mark and the letters of Paul are still in preparation.²¹ The 29th Nestle-Aland edition is announced but not dated.²²
3. The 1 Peter 4:16 flow result depends on settings. It was rerun on the public 2013 data, not on the editors' current database. At connectivity twenty, six "name" witnesses lack a same-reading ancestor within the search limit; at fifty only Vaticanus remains, and its particular link restates the local choice.²⁰ These rank limits are not measured copying-error rates.
4. The origin of the Byzantine text is unresolved. The exact calculation supports a majority presumption under low-error, equal-copying assumptions. The force of a cumulative argument from many readings also depends on shared copying and survival. A common family tree alone does not settle that issue. This article does not establish which transmission history best explains the whole Byzantine text.
5. Papyrus dates are disputed. This article gives the Münster register's dates and notes Nongbri's case for a fourth-century P75. If P75 is fourth century, the early-third-century anchor for the Vaticanus-type text weakens.
6. Sources not fully read. The 1994 second edition and the new commentary accompanying the 2025 edition of the United Bible Societies text were not read, so all committee ratings here are the 1971 ratings and definitions. The Mark 1 and Revelation 22 pages of the 2025 edition were not in the publisher's preview, so those readings were checked in the major edition's online apparatus instead. Primary works by Peter Ruckman and Gail Riplinger were not obtained. The Trinitarian Bible Society's formal statement of doctrine was not obtained, only its published article. Jan Krans's 2011 critique of Thomas Holland on Revelation 22:19 was not obtained. No primary Byzantine-priority argument at Mark 1:2 was read. The patristic citations at John 1:18 are Ehrman's report. Minuscule 1739 at Romans 5:1 could not be checked on images, and the Münster transcription and Metzger disagree about it. Manuscripts 0243, 1739 and 424 at Hebrews 2:9 rest on Metzger's list.
7. The probability models are idealized. They use the Hodges assumptions or deliberately simple variants of them. They show what those assumptions imply. They are not estimates of the real tradition.

Verdict

Textual criticism is a historical inference, not a vote.  Every school examined here, from Westcott and Hort to Robinson's Byzantine priority and Hodges and Farstad's Majority Text, agrees in principle that copies must be grouped by descent where that can be shown, and that a reading's age, its support and its fit with the author all need explaining.  The seven cases show that no single slogan decides.  The oldest manuscripts lose at Romans 5:1 and 1 Peter 4:16, both times to the Byzantine reading.  The Byzantine majority loses at Mark 1:2, on the age and spread of the rival reading.  The shorter-reading rule fails for the early scribes Royse measured.  John 1:18, Hebrews 2:9 and 2 Peter 3:10 remain open.  Revelation 22:19 is not a test of the majority at all, but of the printed Textus Receptus against the whole early Greek tradition.

These cases cannot rank the methods across the New Testament.  They show narrower things.  The critical editions give checkable reasons, mark their doubts, and have moved toward the Byzantine text where their evidence pointed.  Their genealogical checks are real but depend on settings and on the editors' own prior decisions.  Byzantine priority is a coherent rival.  The early witnesses to individual Byzantine readings and the critical editors' reappraisal give reasons to take it seriously;  neither establishes its history for the whole text.  The majority calculation supports a presumption under careful, equal copying.  Its application to real transmission and to many readings together needs further historical evidence.  The Textus Receptus positions rest first on a doctrine of providence that manuscripts cannot test in general.  Where their defenders make specific factual claims, the Trinitarian Bible Society's "nearly consistent with the majority" is roughly right, manuscript 141 fails as a Greek source for "book of life," and "no single Received Text" is their own concession.  The possibility of another pre-Erasmus Greek witness has not been ruled out.

The uncertainty that can be measured is located at known places among the surviving witnesses, and the editions label it.  In the Catholic Letters the editors leave 43 places undecided out of about 3,046 where the copies differ, and only at 2 Peter 3:10 do they print a reading no Greek manuscript has.  These counts show only the doubts that left a mark.  The uncertainty is not nothing, and the right response is to say so verse by verse rather than to manufacture certainty in either direction.

Sources