Their own tradition names the date: about 175 BCE, seven men and seven women, the sole survivors of a shipwreck off the Konkan coast south of Bombay, fleeing the persecution of Antiochus Epiphanes. For two thousand years the Bene Israel of western India kept the Sabbath, circumcision and the Shema while losing everything else, Hebrew, the Torah scroll, even the memory of Hanukkah, and were known to their neighbours as Shanivar Teli, the Saturday oil pressers. This article tests what is left of that origin story in the genome. The answer is a population that is genuinely admixed, roughly six in ten Indian and four in ten Middle Eastern, in numbers that match an independent 2016 study almost exactly, and that fails a placebo test the way a real signal should and a coincidence should not.
A legend with a documented double
The shipwreck story is not unique to the Bene Israel, and that is precisely what makes it a useful test case rather than a dismissible fairy tale. The Chitpavan Brahmins of the same Konkan coast tell an almost identical story: fourteen foreigners, seven men and seven women, drowned in a shipwreck on the same stretch of shore, revived by the sage Parashurama and initiated into Brahminical rites. Tudor Parfitt, who has studied both traditions, considers the parallel too exact to be coincidence and reads the Bene Israel version as a local narrative template borrowed rather than an independent memory. That does not make the Jewish ancestry claim false. It does mean the shipwreck itself cannot be treated as evidence, and that whatever the genome shows has to stand on its own.
What is not in dispute is the community's more recent history. Bene Israel villagers pressed oil and did not work on Saturdays, a habit distinctive enough that it became their caste name. Contact with organised Judaism came late and from outside: Cochin Jews visiting Bombay in the eighteenth century, then Baghdadi Jewish merchants in the nineteenth, taught the Bene Israel Hebrew liturgy and reconnected them to normative practice. The community moved to Bombay, Pune and Ahmedabad under British rule, prospered in the army and civil service, and after 1948 the great majority left for Israel. Fewer than five thousand remain in India today.
Data and method
Everything below uses Global25 scaled coordinates, merged first occurrence wins from the published Davidski files and the Moriopoulos 2026 collection, giving 30,076 entries. Ancestry proportions come from non negative least squares with a sum to one constraint imposed by appending a heavily weighted row of ones, the resulting weights normalised. Distances are scaled Euclidean distances multiplied by one thousand, the convention used throughout this site.
The Bene Israel are represented by the Mumbai_Jew population average, six individuals. That is a small sample and every claim below is checked for robustness against source substitution rather than rested on one pole. The local Indian baseline is Marathi, the general population of the Konkan and Deccan region the Bene Israel have lived among for two millennia, with Konkani Catholic and Maratha averages used as alternates. The Middle Eastern side is tested against six different poles: Iraqi, Syrian, Yemenite and Ashkenazi Jews, the Paradesi community of Cochin, and an Iron Age Levantine ancient average, so that no single claim depends on one population being the correct proxy for whichever Jewish community actually reached the Konkan coast.
Two honest notes before any number is used. Six individuals is a thin sample for a population average, and the Bene Israel are known to have gone through a bottleneck and high endogamy, so this average may not be fully representative of the community's internal range. And a Global25 coordinate is an autosomal summary with no dates attached and no way to distinguish maternal from paternal contribution, so nothing below speaks to sex bias or to admixture timing, both of which come from published work using other methods.
Where the Bene Israel actually sit
The simplest measurement first.
Global25 distance from the Bene Israel to every relevant population, ancient and modern. Orange marks Jewish and Levantine populations.
The single closest population in the entire panel is the Parsis of India, at 21.9 units, a genuine parallel rather than an accident: a Middle Eastern-origin community, Zoroastrian rather than Jewish, that arrived on the same Gujarat and Konkan coast centuries ago and underwent the same process of admixture with the same local population. Behind the Parsis come Gujarati and Punjabi Muslim populations with their own histories of West and Central Asian paternal ancestry, then Punjabi Brahmins and Jat Sikhs, high-ANI Hindu populations with no claimed Middle Eastern ancestry at all.
No Jewish population appears until rank six, the Cochin Jews at 80.0. Every other Jewish or Levantine population, Iraqi, Syrian, Ashkenazi, Iron Age Israel, Yemenite, sits between 200 and 250, further from the Bene Israel than the Konkan's own tribal Warli population at 196.3. Read on its own this ranking would suggest no Jewish connection at all: the Bene Israel look overwhelmingly like an Indian population, and specifically like the class of Indian and Pakistani populations that carry an unusually high ANI, west Eurasian-related component. That is exactly the ambiguity the next section is built to resolve.
The two-source model, and why it has to be tested against itself
A population that carries high ANI ancestry will sit closer to Middle Eastern populations than a population with more ASI ancestry does, for reasons that have nothing to do with any Jewish history whatsoever. So before any Jewish percentage is quoted, the model has to be checked against exactly the failure mode that would produce one by accident: could any high-ANI Hindu caste be made to return a similar number, purely because the Middle Eastern pole happens to sit in roughly the right direction on the ANI axis?
Run the straightforward model first. Bene Israel as a mixture of Marathi, the local baseline, and a single Middle Eastern Jewish source.
Two-source NNLS models, Bene Israel as Marathi plus one Middle Eastern or Levantine source, with the source substituted six ways. Condition numbers 1.2 to 1.7 throughout, no collinearity trap.
The Jewish share lands between 37 and 44 per cent no matter which Middle Eastern population stands in for it, whether that is Iraqi Jews, Syrian Jews, Ashkenazi Jews, the Paradesi community of Cochin, or an Iron Age Levantine ancient average with no Jewish label at all. That range is a strong internal check on its own. It is a stronger one once set against the only independent estimate in the literature: Waldman and colleagues, using GLOBETROTTER on a genome-wide SNP panel entirely unrelated to Global25, put the Bene Israel at 55 per cent Indian and 45 per cent Jewish. A completely different method, run on different markers by a different team nine years earlier, lands inside two points of the Global25 estimate.
The placebo test
Agreement with a published figure is reassuring but it does not by itself rule out the collinearity problem, because the published estimate could in principle suffer from the same confound. So the decisive test has to be run inside this dataset: take the identical Marathi plus Iraqi Jew model and point it at Hindu populations from the same broad region who carry high ANI ancestry and have never claimed a drop of Jewish ancestry, and see what the model does to them.
Residual of the Marathi plus Iraqi Jew model. Bene Israel fits the model almost exactly. The controls do not, whatever weight the model assigns.
The controls do draw a non-trivial weight onto the Iraqi Jewish source, between 20 and 41 per cent, because a Middle Eastern population is the best available proxy in this two-source system for generic ANI-related ancestry the model has no better way to express. That is the collinearity risk stated plainly rather than hidden. But the weight is where the resemblance to Bene Israel ends. The residual, which measures how well the two chosen sources actually reconstruct the target, stays large for every control: 45.4 for the Chitpavan Brahmins who share the shipwreck legend, 61.4 for Gujarati Brahmins, 66.0 for Punjabi Brahmins, 83.1 for Punjabi Jat Sikhs. Bene Israel's residual is 18.6, two and a half to four and a half times smaller. The model is not merely willing to assign Jewish-labelled weight to Bene Israel. It is able to reconstruct Bene Israel almost exactly using that weight, in a way it cannot do for any of the four Hindu populations tested, regardless of how much weight it is given to work with.
That distinction, between weight assigned and fit achieved, is what separates a real shared-ancestry signal from an artefact of a shared regional axis. A confound produces a plausible-looking percentage on a population that does not actually fit the model. A genuine signal produces both.
What this method cannot say
Two limits, stated before anyone else states them. A Global25 coordinate cannot date an admixture event or tell paternal from maternal ancestry. The dating in this article, 19 to 33 generations ago by two independent methods, ALDER and GLOBETROTTER, roughly six to eleven centuries depending on the generation length assumed, comes entirely from Waldman and colleagues 2016, not from anything computed here. The same paper found the admixture was sex-biased: Y-chromosome markers, including the Cohen Modal Haplotype common among Jewish priestly lineages, point toward a Middle Eastern paternal contribution, while mitochondrial DNA is overwhelmingly of local Indian origin. More Jewish fathers, more Indian mothers. Global25 cannot see chromosomes separately and cannot reproduce that finding; it is reported here because it is the single most informative fact this article does not itself establish, and because it explains why the autosomal signal above sits at a real but modest 37 to 44 per cent rather than anything higher: six or so centuries of endogamous marriage inside a small community dilutes an initial founding contribution without erasing it, exactly as the two-source model finds.
Myth and measurement
| The tradition | What the genome records |
|---|---|
| Fourteen shipwreck survivors from northern Palestine, around 175 BCE, is a specific and checkable founding event. | The same story, survivor count included, is also the origin legend of the Chitpavan Brahmins of the identical coast. The shipwreck cannot be used as evidence either way. |
| Bene Israel autosomal ancestry looks entirely Indian, which some earlier studies read as no evidence of Jewish origin at all. | Raw distance alone supports that reading: no Jewish population appears among the ten nearest neighbours. A two-source model is needed to recover the signal at all. |
| Any high-ANI Hindu population could in principle be made to show a similar Middle Eastern Jewish percentage, which would make the whole exercise meaningless. | Tested directly against four such populations. They draw real but smaller weight onto the same source and fit the model two and a half to four and a half times worse than Bene Israel does. |
| A Jewish contribution, if real, should be a small trace after two thousand years. | It is not a trace. Six different Middle Eastern poles agree on 37 to 44 per cent, matching an independent 2016 estimate of 45 per cent almost exactly. |
The coordinates
Global25 scaled coordinates for the populations used above, for anyone who wants to reproduce the models in Vahaduo.
Mumbai_Jew_(n=6),0.062982,0.006601,-0.119799,0.042259,-0.055805,0.028586,0.000666,0.006115,0.020146,0.014883,-0.001191,-0.001474,0.003593,0.004519,-0.003936,-0.002674,-0.005215,0.001605,0.002221,-0.003397,0.000541,-0.003256,0.000658,0.003976,-0.002495
Marathi_(n=7),0.035773,-0.093429,-0.169650,0.083517,-0.052147,0.041307,-0.002619,0.011169,0.041014,0.026886,-0.002396,-0.000984,0.003013,0.004123,-0.005247,-0.003331,-0.004143,0.001102,0.001923,-0.002986,0.001358,-0.004125,0.000913,0.003791,-0.002891
Cochin_Jew_(n=21),0.046830,-0.046182,-0.146808,0.084949,-0.067983,0.049656,-0.001276,0.008670,0.030688,0.018605,-0.001694,0.000014,0.001104,0.001016,-0.002734,-0.002008,-0.000062,-0.001074,0.000545,-0.002096,0.002751,-0.003674,0.000012,0.001148,-0.002241
Iraqi_Jew_(n=11),0.091369,0.138574,-0.065893,-0.075376,-0.025263,-0.022337,-0.000491,-0.008139,-0.010524,-0.001441,0.003956,-0.005967,0.012325,0.005005,-0.009266,0.009390,-0.005713,0.000910,0.002388,-0.009777,0.001940,-0.002169,-0.003171,-0.002213,0.001644
Parsi_India_o,0.067535,0.003385,-0.119547,0.043800,-0.058200,0.030100,-0.001100,0.005900,0.019700,0.014200,-0.005100,-0.000700,0.004300,0.003100,0.000500,-0.005300,-0.009800,-0.000200,0.000600,-0.003200,-0.001400,0.001500,References
-0.000500,0.005700,-0.002900
- 2016 Waldman YY, Biddanda A, Davidson NR, Billing-Ross P, Dubrovsky M, et al. "The Genetics of Bene Israel from India Reveals Both Substantial Jewish and Indian Ancestry." PLOS ONE 11:e0152056. Read the paper