Around the year 207, Tertullian worked through Marcion’s Gospel in order, answering it passage by passage in the fourth book of Against Marcion. When he reaches the fifteenth chapter of Luke, he takes up the lost sheep and the lost coin, and then, for twenty-two verses, he says nothing at all. The parable of the prodigal son, one of the best-loved stories in the church’s Gospel, goes by without a word, and he resumes only with the dishonest steward in the next chapter. Some 170 years later Epiphanius, copying excerpts from his own copy of Marcion’s Gospel into his Panarion, supplies the reason: the parable was not there.
Tertullian’s silence and Epiphanius’s report fit together, and they are far from the only pair that does. Every reconstruction of Marcion’s Gospel, from Harnack’s to the editions published in the last decade, has to decide what to do wherever its witnesses say nothing, and that problem covers more of the text than most readers realize. Our witness tables, built for the Greek True Testament Critical Edition and published on our Open Data page, record every statement that Tertullian, Epiphanius and the anonymous Dialogue of Adamantius make about the text of Marcion’s Luke. Once the quotations that come from the writers’ own Bibles are set aside, along with Book V of the Dialogue, where no Marcionite speaks and which we do not count as a witness to Marcion’s text except for the one Gospel passage its speaker reads “from their own gospel” (Luke 18:35–43), the witnesses report 529 verses of Luke as present in Marcion’s Gospel and 275 as missing, and they disagree about three. On the remaining 342 verses, nearly a third of Luke, no witness says anything at all.
We set out to measure what those silences are worth. This article reports what we found, including one test that did not hold up when we checked it properly: how Tertullian’s silences behave, what they do and do not show, the model we built from them, the checks we ran on it, what it suggests about the length of Marcion’s Gospel, how the five modern reconstructions, our own among them, line up with it, and where the model is weakest. Every number can be checked verse by verse on the new research page, Weighing the Silences.
The Pattern in Tertullian’s Silences
The witnesses do very different jobs. Tertullian reports far more of Luke than anyone else, but beyond the opening of the Gospel he almost never says that a passage was missing. Epiphanius works the other way: he quotes much less, but he names passage after passage that Marcion’s text did not have. Beyond the opening chapters he reports 93 such verses, and Tertullian confirms none of them; he is silent on 91 and reports the other two as present in Marcion’s text.
Those two facts fit together, and the fit is the strongest result in this study. Tertullian reports about half of Luke from Marcion’s text, quoting or paraphrasing it. From the genealogy on, Epiphanius reports 111 verses missing, and if the two writers had nothing to do with each other, Tertullian would have reported about 54 of them as present. He reports two. Whatever else is uncertain, the two witnesses describe the same text.
Tertullian’s commentary leaves 129 stretches of silence between the verses he reports. Most are short, typically two verses, which is what we would expect from a writer passing over lines that do not serve his argument. The stretches that contain a passage Epiphanius reports missing are longer, typically nine verses, and 64 of the 91 verses Epiphanius reports missing sit inside silences of ten verses or more. When this article first appeared, it said that a difference that large never appeared once in twenty thousand random shuffles. That test was flawed: it ignored that a long missing passage can only fit inside a long silence. When Epiphanius’s reported passages are kept at their real sizes and placed at random inside Tertullian’s silences, they land in silences just as long as the real ones, about fifteen verses on average either way. The long silences around Epiphanius’s reports are what the size of those passages forces. By themselves, they do not show that a long silence is a sign of absence.
In the clearest cases the two witnesses match almost verse for verse. Tertullian is silent for 22 verses across the prodigal son, exactly the 22 verses Epiphanius says Marcion’s text did not have. He is silent for the first nine verses of Luke 13, the Galileans killed by Pilate, the tower of Siloam and the barren fig tree, and those are exactly the nine verses Epiphanius reports missing. He is silent for eleven verses around the parable of the tenants, nine of which Epiphanius reports missing. Two writers working nearly two centuries apart, with different aims, point to the same places.
From a Pattern to a Probability
The model turns that pattern into a probability for every verse. Inside Tertullian’s silences there are two kinds of verse whose fate we know: the verses Epiphanius reports missing, and the verses Epiphanius or the Dialogue quotes as present. Comparing the two shows how the length of a silence, and a verse’s place within it, goes with an absence among the verses someone reports. The model then assumes that the same holds where no one reports anything. That assumption is its central step, and as the previous section showed, the witnesses do not prove it. Every probability it gives for a verse no witness mentions rests on that assumption.
One quantity still has to be estimated: how many of the passages missing from his copy Epiphanius actually recorded. The data answer that too, by comparing how often the witnesses quote verses that were present with how often Epiphanius reports verses that were not. His thoroughness comes out at about 44 percent, with a likely range from 21 to 67 percent. Every result below is given with that uncertainty, and the research page lets a reader see how each verse moves if Epiphanius recorded a quarter or three quarters of the missing passages instead.
We also tried a more ambitious version, which modeled every silence as Tertullian skipping some verses and Marcion’s text lacking others, and it failed. The witnesses as a group go quiet in silences of four to nine verses, quoting less and reporting less, and a model built on raw rates read that general quiet as absence. The design we kept compares one kind of report against the other inside the same silences, which is why it does not make that mistake.
Building the model tested our own data as well. It showed that the witness tables on our Open Data page lacked several reports the Critical Edition already relied on, among them Epiphanius’s statement that Marcion’s Gospel had no account of the births, the baptism or the genealogy, and nineteen verses that Tertullian or Epiphanius quotes from Marcion’s text. Those reports are now in the tables, and every result here is computed from the corrected data.
Four Checks on the Model
We checked the model in four ways. The first three are tests it could fail, run on evidence it had not been shown, and it passed them. The fourth checks an assumption the data cannot settle. All four grade the model against Epiphanius’s reports and Tertullian’s quotations, the same kind of evidence it learned from, so they show that it handles that evidence consistently. They cannot show that the length of a silence means anything by itself.
- Finding absences it was not shown. With each silence held out in turn, the model marks 82 percent of the verses Epiphanius reports missing as more likely absent than present.
- Finding presences it was not shown. With Tertullian’s reports hidden in randomly chosen stretches, the model still scores 93, 82 and 78 percent of the hidden verses that no other witness reports as present, for stretches of five, ten and twenty verses, so a long silence alone does not make it declare a passage absent.
- Numbers that mean what they say. On verses held out from the model, those it scored about 33 percent likely to be absent turned out to be absent 29 percent of the time; at 46 percent, 54; at 74 percent, 76; at 87 percent, 84. When the model says a verse is likely absent, it is about as often right as it claims.
- Epiphanius’s own interests. The model assumes Epiphanius was as likely to note a short missing passage as a long one. If he noticed large missing passages more readily, the absences he recorded would make long silences look more telling than they are, and the estimates below would fall by about a hundred verses. The data cannot settle how strong such a bias was.
How Long Was Marcion’s Gospel?
Adding up the probabilities verse by verse gives an estimate of the length of Marcion’s Gospel: about 740 of Luke’s 1,149 verses, with a likely range from 661 to 799, if the model’s assumption about silences holds. If Epiphanius did favor large passages, the figure falls to between about 605 and 645. Because the assumption is not proven, these figures are best read as what follows from it, not as a measurement.
For comparison, Klinghardt prints 797 verses of Luke, Litwa 627, the True Testament 563, BeDuhn 537 and Roth 488. The figures do not all claim the same thing. Litwa, BeDuhn and Roth leave hundreds of verses undecided, and the True Testament prints only what is attested or probable: 251 verses that it does not include are graded “absence not established,” which means that no witness shows them absent and none shows them present. The model suggests that most of those verses stood in Marcion’s Gospel, and the True Testament has never claimed otherwise. It prints what the witnesses allow it to print.
How the Reconstructions Line Up
Because the model gives every verse a probability, each edition’s decisions can be set beside it. For every verse an edition decides, printing it or not, the model gives a chance that the decision is right, and we average those chances. Verses an edition leaves undecided are not scored, so the comparison also has to say how many verses each edition actually decides. Bilby’s reconstruction covers the letters, not the Gospel, so it is not included.
Where a witness reports the verse, the model agrees with every edition between 93 and 96 percent of the time. That is expected and says little: the model mostly repeats the witnesses there, and every edition, the True Testament included, built its decisions from the same reports.
Where no witness speaks, the editions differ first in how many verses they decide at all. Klinghardt decides all 342, and the model agrees with his calls 60 percent of the time. Litwa decides 143 (63 percent), the True Testament 119 (55 percent), BeDuhn 57 (65 percent) and Roth only 17 (75 percent). These figures measure agreement with the model, not with any evidence, and those built on few verses rest on very little. The True Testament decides these verses on grounds the model cannot see, such as the argument a passage belongs to, which is why the model agrees with it least here. That is a difference to examine, not a verdict.
Where the Model and Our Edition Disagree
We asked the model the hardest question we could: where does it disagree with the Greek True Testament Critical Edition? At first it flagged 25 verses that no witness mentions. We reviewed every one against the Latin and Greek, and the review led to real changes. Five passages that the Critical Edition had graded as probable absences rested only on the inferences of Harnack and Zahn, with no witness behind them, and they now stand as “absence not established”: Luke 5:39, 9:27, 10:12–15, 24:22–24 and 24:45–46. Two verses moved from attested to probable. None of these changes alters the text of the True Testament, and none of those passages is included in it.
Eight disagreements remain, and in each the Critical Edition’s grade rests on evidence the model cannot weigh. Tertullian says Jesus was thrown out of Nazareth for one proverb, and Ephrem says he spoke nothing but that one saying, which stands against Luke 4:24–26. Epiphanius reports Luke 12:6 missing, and 12:7 finishes its saying about the sparrows; he reports 22:16 missing, and Tertullian goes from 22:15 straight to the bread, past 22:17–18. Tertullian quotes the answer of Luke 20:8, which turns on the question of 20:2. For Peter’s denial at 22:57, Tertullian reports Jesus’ prediction of it, and Epiphanius, who quotes 22:47 and 22:63–64, reports nothing missing between them except 22:50. A model that sees only reports and silences cannot see an argument from wording or order, and these eight are where that limit shows.
Five Silences Still Unexplained
Beyond the opening of the Gospel, Tertullian falls silent for ten verses or more thirteen times. Eight of those silences contain passages Epiphanius reports missing. In the other five, no witness reports anything missing at all:
- Luke 6:47 to 7:8, the end of the sermon and the opening of the centurion’s story (11 verses).
- Luke 8:33 to 8:42, the end of the Gerasene story and the beginning of Jairus’s (10 verses), where Epiphanius quotes only 8:42.
- Luke 10:28 to 10:42, the parable of the Good Samaritan and the visit to Mary and Martha (15 verses), where Epiphanius quotes only the opening line, 10:28.
- Luke 14:27 to 15:3, the cost of discipleship and the opening of the lost sheep (12 verses).
- Luke 22:23 to 22:32, the dispute about greatness and the words to Simon at the Supper (10 verses).
The model treats these as genuinely uncertain, and so do we. Some of them may mark absences that Epiphanius did not record, and some may be passages Tertullian simply passed over. They are where the next piece of evidence, if it is ever found, will matter most.
Where the Model Is Weakest
The model has real limits, and they belong beside its results. The most serious is the one described above: it reads a long silence where no one reports anything as a sign of absence, and the witnesses do not prove that. Its second assumption, that Epiphanius was as likely to note a short missing passage as a long one, carries a lot of weight, and Epiphanius was looking for passages he could use against Marcion, so his notes are probably not a random sample of what was missing. The verses it learns from were not chosen at random either: it learns what a verse that stood looks like from verses Epiphanius or the Dialogue quotes, and they quoted what served their argument. It has little to learn from, only thirteen long silences and a limited number of blocks of Epiphanius’s reports, which is why the estimate of his thoroughness runs from 21 to 67 percent. And it sees only the length of a silence and a verse’s place in it, never what a passage says or whether Tertullian had a reason to mention it.
What a Silence Is Worth
Our Systematic Theology states the principle we have always held: a silence “agrees with an absence; it does not attest one.” This study bears that principle out. The model never set a grade by itself: every change it prompted was made only after the witnesses had been read again, and the grades of the Critical Edition do not depend on it. What the witnesses do establish is that Tertullian and Epiphanius describe the same text. What the model adds is a way to state, verse by verse, what follows if a long silence does point to an absence, together with a check showing that the witnesses alone do not prove it does.
The full results are open to anyone who wants to check them. The research page shows every verse of Luke with its probability, the reports of each witness and the decisions of each edition, and the Open Data page offers the same table for download, along with the witness tables it was built from. Silence remains the weakest kind of evidence in the study of Marcion’s Gospel and the most common, and the honest result of trying to weigh it is to know how little it can carry on its own.
Updated 8 November 2026. This article now reflects version 1.2 of the model. It counts the one Gospel passage in Book V of the Dialogue that the Critical Edition accepts, and it scores each edition only on the verses that edition decides; the first version counted an undecided verse as a fifty-fifty call, which favored editions that leave many verses open. It also withdraws the claim that the link between Tertullian’s long silences and Epiphanius’s reports could not have arisen by chance. That test ignored the size of the missing passages, and once it is allowed for, the link disappears.