I grew up in a community where you didn't read certain books. Not banned the way secular novels were frowned upon — older than that, quieter than that, a prohibition that didn't need to be spoken aloud to be obeyed. The Gospels were among those books. So were Paul's letters. I knew they existed. I knew half the world around me quoted them, argued about them, defended against them, almost always without having read them either. When I finally left that world, my first priority was straightforward: find out exactly what I'd been told to fear.
I didn't buy a Spanish or French translation. I bought Franz Delitzsch's Hebrew translation of the New Testament — the nineteenth-century Hebraist's version, in its modern Negev Version revision — specifically because I wanted to meet the text in the language I actually think in, without a foreign-language buffer. I opened it to Romans 9. And I'll be honest, because honesty is the only standard worth holding here: my pulse changed while I turned the pages. Not from fear of the forbidden. From what I was finding.
Five citations from the Torah, and all five slotted into the Masoretic Hebrew as though they'd never left it. For a moment, I thought I'd been wrong about something much bigger than I'd suspected.
What actually unseated that impression wasn't philology. It was remembering something I already knew about the book in my hands.
Delitzsch's translation wasn't produced in a neutral seminar. It was commissioned and funded by nineteenth-century Protestant missionary societies with one explicit goal, which Delitzsch never hid: convert Hebrew-thinking Jews by handing them a New Testament that read, line for line, like their own Bible. As a conversion tool it failed — printed, revised, recirculated for a century and a half, and it never once achieved what it was built to achieve. What it did achieve, unintentionally, took me years to understand properly: the more fluent and "biblical" a Hebrew text sounds when presented as a translation, the more suspicious it deserves to be, not less. A skilled translator working Greek-to-Hebrew has a luxury Paul never had going Hebrew-to-Greek — he can simply return the citation to its original Masoretic form, with no scar tissue showing where he had to travel to get there.
That's exactly what Delitzsch does. Where Paul's Greek fuses two separate Genesis verses into one impossible citation, using a verb that appears in neither of them, Delitzsch just writes the correct, singular Hebrew form and the seam vanishes without trace. His Hebrew is good precisely because it's faithful to the Hebrew Bible — which means it systematically hides every fracture that's visible in Paul's Greek to anyone who knows where to look. I closed Delitzsch that night with more questions than I'd started with, and I began, for the first time, comparing.
The claim, stated fairly
There's a line Hebrew Roots and Messianic apologists reach for whenever this discussion threatens to get uncomfortable: Paul was a Pharisee, son of a Pharisee, trained at the feet of Rabban Gamaliel the Elder — grandson of Hillel, the most venerated legal authority in first-century Jerusalem. Acts 22:3 puts it in Paul's own mouth. The implication is simple: if Paul's Torah citations look strange, it's not because he misunderstood the text. It's because he understood it with a rabbinic authority a twenty-first-century Jew is supposedly missing.
I wanted to believe that too, for a while. Then I started laying his Romans 9 citations — as the Codex Sinaiticus preserves them, folio 264b, our oldest complete New Testament manuscript — next to the Hebrew. Then next to the Aramaic Targumim. Then next to a witness almost nobody uses for this: the Judeo-Greek Constantinople Pentateuch, printed in 1547 by Eliezer ben Gershom Soncino, the written record of a continuous Greek-speaking Jewish translation tradition stretching back to the Hellenistic period. What I found, verse after verse, isn't a rabbinic authority exercising textual expertise. It's something much closer to a man who learned Scripture by ear, in Greek, and never once in the entire chapter goes back to the Hebrew to check whether the Greek he's citing says what he needs it to say.
Here's the method: five witnesses in parallel for each citation — the Masoretic Hebrew, the Septuagint, Sinaiticus for Paul's actual wording, Targum Onkelos and Targum Yonatan, and the Constantinople Pentateuch. Plus a sixth witness that appears only twice, but decisively: Papyrus 46, dated to 175–225 CE — a century and a half before Sinaiticus, and less than a human lifetime from Paul's own death. Where P46 confirms a reading, I'm looking at something as close to the text Paul's first readers actually held as textual criticism will ever let me get.
Case one: the citation that fuses two verses into an impossible third
Romans 9:9 gives us: "at this time I will come, and Sarah shall have a son." The problem is that this sentence exists nowhere in Genesis. It's a collage. "At this time" comes from Genesis 18:10. "Sarah shall have a son" comes, almost word for word, from Genesis 18:14. And the verb stitching them together — eleusomai, "I will come" — appears in neither Septuagint verse: 18:10 has "I will arrive," 18:14 has "I will return," with a prefix that explicitly carries the sense of coming back. Paul isn't quoting a text. He's remembering a tune and humming whatever lyric comes out, with a third word that isn't in either score.
Every Jewish translator I checked — Onkelos, Targum Yonatan, the Constantinople Pentateuch, and, independently, the twentieth-century Hebrew–Spanish translation by Moisés Katznelson, produced for Spanish-speaking Jewish communities with zero interest in the New Testament either way — keeps 18:10 and 18:14 as two distinct announcements, each with its own verb, its own tense. Nobody, in any century, with the Hebrew in front of them, produces the fusion Paul produces. And this isn't a fourth-century copying error: Papyrus 46 has the identical fused reading, a century and a half before Sinaiticus. Either Paul dictated it that way from memory, uncollated against either source verse, or the error entered the tradition so early it makes no practical difference which.
Case two: the verb Paul didn't need to borrow
Genesis 21:12, cited in Romans 9:7 as "through Isaac shall your offspring be called," uses the Greek verb kaleō — "to call, to name." Paul doesn't let it go: it's the verbal hinge his entire argument about divine election in the chapter hangs from, recurring at 8:30 and again at 9:24.
The Hebrew behind the citation is qara, and qara doesn't primarily mean "to designate." It means to proclaim, to cry out. The Septuagint's kaleō is a defensible rendering in context — but it's a choice, not a necessity. Every independent Jewish witness makes a different one. Onkelos uses the passive of the same root, no drift toward "designate." The Constantinople Pentateuch translates with a verb built on krazō — "to cry out," the exact phonetic and semantic cognate of qara. Katznelson, working in the twentieth century with zero apologetic interest either way, renders it as a verb of growth and expansion — further still from "naming" than the Constantinople Pentateuch. Three independent Jewish traditions, two language families, a millennium and a half apart, and none of them needed kaleō. Only Paul needed it — because without it, he loses the verbal hook his entire doctrine of calling depends on.
Case three: the hierarchy the Greek translator imposed
Genesis 25:23, cited in Romans 9:12 as "the elder shall serve the younger," uses Greek comparatives — meizōn, elassōn — carrying a formal rank-hierarchy that classical Greek reserves for status relationships. That's the Septuagint's choice, and Paul reproduces it untouched. But the Hebrew, rav ya'avod tza'ir, uses plain positive-degree adjectives, no comparative apparatus at all. The Constantinople Pentateuch renders it in the vernacular positive degree — "the big one serves the small one" — no formal hierarchy imposed. Katznelson, independently, five centuries later: "the elder shall prevail," not "shall outrank." Paul inherits, unexamined, a rhetorical weight a Greek translator imposed three centuries before he was born, and builds his argument about unconditional divine election on that weight rather than on the Hebrew itself.
Case four: the doublet the Septuagint flattened
Exodus 33:19, in Romans 9:15, is where Paul is actually faithful to his source — word for word accurate to the Septuagint — which makes this case different in kind from the others: not a slip of memory, but blind fidelity to someone else's error. The Hebrew has two genuinely distinct roots: chanan (unearned favour) and racham (visceral compassion, same root as "womb"). The Septuagint flattens both under near-synonymous Greek verbs. Onkelos preserves the two roots intact. Targum Yonatan does too, with a separate pair, and adds something theologically load-bearing that Paul never cites: the targum explicitly introduces a criterion of perceived merit — "I will show mercy on him whom I see fit for mercy" — which is exactly the contextually correct reading, since Exodus 33 is Moses interceding after the Golden Calf, a dialogue of plea and response, not an abstract predestination treatise. The Constantinople Pentateuch, independently, uses two genuinely distinct Greek verbs to preserve the same distinction the Hebrew marks. Four independent Jewish witnesses converge on keeping a distinction that only the Septuagint — and, by inheritance, Paul — erases.
Case five: the exception that almost broke the argument
This is the one that nearly made me throw the whole piece out, because for a moment it seemed to contradict everything above. Exodus 9:16, in Romans 9:17, has Paul using an active verb — "for this very purpose I raised you up" — where the standard Septuagint has a passive: "you were preserved." My first instinct was alarm: if Paul departs from the Septuagint in exactly the direction his theology needs, am I the one being tendentious here, seeing error where there's only useful rhetorical coincidence?
The alarm settled when I confirmed I'm not the first to flag the divergence — Catholic exegete Senén Vidal notes the same textual friction in passing, with no polemical stake in the observation at all. And P46 has the identical active reading, ruling out the comfortable explanation that a later, more careful Christian scribe corrected the text toward the Hebrew over the centuries. So I went to the Hebrew, and there's the real surprise: the causative Hifil form of the Hebrew verb genuinely is active — "I made you stand, I raised you up." On this one point, out of five, the standard Septuagint is further from the Hebrew than Paul's citation is. Onkelos confirms it independently, active voice; Targum Yonatan too, with its own polemical addition explicitly wrestling with the same question of Pharaoh's treatment a thousand years before I did.
I have to be straight about the rest of it, though, because showing the complicating evidence matters more than protecting the thesis. Katznelson's modern Hebrew–Spanish translation renders the same verb as "I let you live" — closer to the Septuagint's sense of passive preservation than to the Targumim's active erection. The Hebrew causative genuinely admits both readings, and not every Jewish witness lands in the same place. What they agree on is that God is the one acting on Pharaoh, not that Pharaoh is a passive object of anonymous preservation, as the Greek syntax has it. That's a real, narrower point than "every witness says active" — and I'd rather flag the softening myself than have someone else find it first.
A witness that changes nothing about the pattern, and I want to be honest about that too
The oldest Hebrew witness available, the Qumran scroll 4Q1, doesn't preserve any of these five verses — its surviving fragments simply don't reach that far. What it does show, in the passages where it survives, is a consonantal text sticking closely to what would become the Masoretic tradition, against the more expansive, harmonising Samaritan Pentateuch — down to details no scribe would have a motive to invent. It also shows, at Exodus 1:5, a genuine textual variant (seventy-five descendants rather than seventy) that agrees with the Septuagint rather than the Masoretic Text — proof that real textual plurality existed in the Second Temple period, and that the Septuagint isn't always simply "wrong." But that variant is a difference in the underlying Hebrew manuscript the translator had in front of him. None of the five divergences documented above are of that type. Every one of them depends not on which Hebrew text the Alexandrian translator had, but on what he — and later, Paul — did with a consonantal text that is, at every point that matters here, the same one preserved by the Masoretes, the Targumim, and the Constantinople Pentateuch alike.
The credential itself doesn't hold up
Here's where this gets uncomfortable for reasons that have nothing to do with grammar. Acts has an internal problem the "Paul the rabbinically trained Pharisee" reading has to explain away: when Peter and his companions are brought before the Sanhedrin, it's Gamaliel himself — Paul's supposed teacher — who stands to defend them, comparing the Nazarene movement to others that will simply dissolve if they aren't of God. The entire Pharisaic bloc votes with him. As the Jewish scholar Hyam Maccoby pointed out in a 1986 lecture, the only Pharisee in the whole of Acts documented persecuting the Nazarenes is Paul — the Pharisees aren't cast as persecutors anywhere else in the record, even in its more hostile tellings; that role belongs consistently to the Roman-appointed Sadducee high-priestly circle. And it's precisely the high priest Paul cooperates with, requesting arrest warrants for believers in Damascus. Maccoby's conclusion: Paul's persecution reads more plausibly as Sadducee-Roman political policing of insubordination than Pharisaic religious zeal — and the Pharisaic credential became convenient only later, writing to Gentile audiences with no way to check the internal politics of Judea.
I want to flag clearly that Maccoby goes further than this, into territory he himself calls a minority position, not a consensus finding: that Paul may not have been born Jewish at all, given his letters never mention Gamaliel or Tarsus as his birthplace — both facts we only have from Acts, written by someone else — and that Tarsus was a centre of the Attis mystery cult, whose dying-and-rising theology bears a closer resemblance to Pauline soteriology than to Pharisaic Judaism's more world-affirming theology. That's Maccoby's argument, clearly labelled as his and as radical. It deserves to be read as an argument, not smuggled in as fact.
The hostile witness
What's notable is that this suspicion isn't a modern invention. It shows up in the fourth century, in the pen of an adversary citing it only to condemn it: Epiphanius of Salamis, in the Panarion, records a tradition held by the Ebionites — Torah-observant Jewish Christians who venerated Jesus as a human Messiah and regarded Paul as a corrupter of his message. According to that hostile source, the Ebionites claimed Paul was ethnically Greek, son of Greek parents, who came to Jerusalem, wanted to marry a priest's daughter, converted and was circumcised for that reason, and — refused the marriage — turned in anger against circumcision, Sabbath, and Torah.
I want to be precise about how much weight this carries, because honesty means distinguishing types of evidence. Epiphanius isn't confirming the Ebionite tradition — he's citing it to condemn it as the invention of "false apostles." But a hostile witness recording a specific, detailed second-century Jewish-Christian tradition — with details no one would have a motive to invent if they weren't already circulating — is exactly the kind of source a historian doesn't discard just because it's inconvenient. I'm not claiming the Ebionite story is historically accurate in every detail. I'm claiming that a community of Jews who knew about Paul, in the first or second century, didn't take for granted — the way most later Christian scholarship does — that he was straightforwardly Jewish by birth, let alone a Pharisee trained under Gamaliel.
Put the three lines of evidence next to each other
First: the philological analysis of five Pentateuch citations in Romans 9 shows, at no point, independent Hebrew consultation — it shows dependence on an imperfectly remembered Septuagint, patched with fragments of orally absorbed Targumic paraphrase, never checked against the original. Second: the internal logic of Acts itself makes Paul's persecution of the Nazarenes read more plausibly as Sadducee-Roman collaboration than Pharisaic zeal, and the Gamaliel credential appears only in a text written by someone else with an interest in legitimising its protagonist. Third: a specific Jewish-Christian tradition existed, from the first or second century, holding that Paul wasn't Jewish by birth at all. None of these three lines depends on the other two. None needs the others to be right in order to stand on its own. All three point, independently, the same direction.
I'm not going to close the question for you. But here's the one I can't put down: if a man builds the most influential theology in Western history on five Torah citations that no first-century Jewish contemporary of his — Aramaic-speaking or Greek-speaking — would ever have cited that way, what exactly are we reading when we read Romans 9? The culmination of Israel's revelation? Or the Greek of a man who never quite finished learning the language he claimed to be interpreting?
Full sourcing — manuscript collations, targumic apparatus, and the complete Epiphanius text with critical notes: https://www.orajhaemeth.org/2026/08/romanos-9-el-griego-que-delata-pablo.html
Reddit Version: https://www.reddit.com/u/OrajHaEmet/s/qeU82KEGbw