By Emilia Wellesley · Published May 6, 2026 · Updated May 6, 2026
Somewhere on the eastern lip of the Gobi, a few hundred farmers in a Gansu village called Zhelaizhai look, by Han Chinese standards, a touch unusual. A few have light hair. A few have green or hazel eyes. A small museum at the edge of town displays a plaster Roman bust and a column drum that would not be out of place on a film set, and the local tourism board has, for years, traded in a single irresistible claim: that these villagers descend from the survivors of a lost Roman legion.
The story is older than the museum. It belongs to a single hypothesis advanced in 1941 by an American sinologist who read the relevant Chinese chronicle carefully, then asked a question no one had quite asked before. Whether the answer he gave is correct is a different matter, and the question this article asks is the historiographer’s question rather than the tour-guide’s: what would it take for the romance to be true, and does the surviving evidence support it?
The Direct Answer
The Lost Roman Legion in China is the hypothesis, advanced by Homer H. Dubs in 1941 and expanded in 1957, that survivors of Crassus’s legions defeated at the Battle of Carrhae in 53 BCE made their way east, fought as mercenaries at the Battle of Zhizhi in 36 BCE, and were resettled at Liqian county in Gansu. The textual evidence is one ambiguous passage in the Han Shu; the genetic evidence does not support Roman ancestry; the chronology of Liqian’s founding undermines the link.
How a Sinologist Read a Han Annal in 1941
Homer Hasenpflug Dubs (1892–1969), an American sinologist who held the chair of Chinese at Oxford from 1947, spent two decades translating the Annals of the Han Shu by Ban Gu (32–92 CE), the dynastic history that runs from the Qin collapse through Wang Mang’s interregnum [1]. Dubs was the rare Western scholar of his generation fluent enough in classical Chinese to read the Hanshu against the grain, and in 1941 he published, in the American Journal of Philology, a short article called “An Ancient Military Contact Between Romans and Chinese.” A book-length expansion, A Roman City in Ancient China, followed in 1957.
His argument moved between three textual signals and one geographic name. The signals came from Ban Gu’s account of the Han general Chen Tang’s 36 BCE assault on Zhizhi Chanyu, a renegade Xiongnu chieftain who had built a fortress somewhere near the Talas River, in what is now the Kazakhstan-Kyrgyzstan border zone. Chen Tang’s report, preserved in the Hanshu, described the defenders organizing themselves into a “fish-scale formation” behind a double wooden palisade. To Dubs, the formation looked like the Roman testudo, the overlapping-shield wall that legionaries used against missile fire. The double palisade, he suggested, looked like a Roman fortification trick. And the place where Chen Tang’s prisoners were ostensibly settled bore a name that, to a sinologist’s ear, could be read against the Greek-Chinese transliterations attested elsewhere in the period: Liqian.
The Carrhae-to-Talas Bridge
Dubs’s reconstruction needed a bridge of seventeen years and roughly four thousand miles. In 53 BCE, Marcus Licinius Crassus (115–53 BCE), the richest man in late-Republican Rome and a member of the First Triumvirate, marched seven legions across the Euphrates against the Parthian Empire. At Carrhae, in what is now southeastern Turkey, the Parthian general Surena annihilated him. Crassus died in the field. The Roman historian Plutarch (c. 46–119 CE) reported that roughly ten thousand legionaries were taken prisoner and sent to garrison the Parthian frontier at Margiana, near present-day Merv in Turkmenistan [2].
Dubs proposed that some of these prisoners drifted east, were absorbed as mercenaries by Zhizhi, and ended up behind that double palisade in 36 BCE. The chain has the shape of a working hypothesis. It also has the shape of a hypothesis that requires every weak link to hold simultaneously: the prisoners survive a generation in Parthian service; they leave Parthian service alive; they cross hundreds of miles of hostile steppe; they arrive in numbers large enough to drill a recognizable formation; and they happen to be at the one fortress the Han chose to overrun. None of these steps is impossible. None is corroborated by anything outside the Hanshu passage Dubs began with.
Reading the Han Shu Passage Carefully
Ban Gu’s account of Chen Tang and Gan Yanshou’s 36 BCE assault is one of the better-preserved campaign reports of the Western Han, and it is almost certainly closer to a primary source than most surviving battle narratives of the era, drawing on Chen Tang’s own dispatch [3]. The passage describes Zhizhi’s roughly three thousand troops mounting a determined defense from a fortified citadel: queen and concubines firing arrows from the ramparts, ten thousand Kangju cavalry harassing the Han rear, and, in the inner ring, “more than a hundred men” arrayed in a yulin or “fish-scale” pattern.
Three things are worth noting about that line. First, the formation is described, not pictured. The Chinese phrase translated as “fish-scale” is a metaphor with a long internal history in Chinese military writing; it appears in Sun Bin’s Art of War, predating Roman contact by centuries, as a generic image of overlapped tight order. Second, the hundred-odd defenders are not described by ethnicity, dress, or weapons; nothing in Ban Gu’s text says these men looked foreign, spoke a foreign tongue, or carried foreign equipment. Third, the wooden palisades are doubled rather than concentric in the Roman manner, and double palisades are widely attested in Central Asian and Indian fortifications from the period [3]. Yang Gongle of Beijing Normal University, whose 2007 critique remains the standard skeptical reference, makes the point that every signal Dubs read as Roman has at least one well-attested non-Roman precedent. The testudo reading is not impossible. It simply does not earn its keep against the alternatives, and a reading that requires four other improbabilities to also hold is methodologically expensive.
There is also the small matter of what happened to the hundred men. Ban Gu records that 1,518 of Zhizhi’s people died in the assault and roughly a thousand more were captured or surrendered. The chronicle does not single out the fish-scale unit for resettlement, transportation, or any subsequent administrative act. If they were Romans, the Han bureaucracy that lavished careful record-keeping on the war’s accounting did not notice it.
The Chronology Problem
The hardest blow to Dubs’s reconstruction is administrative rather than archaeological. Liqian county, the place where Dubs’s prisoners were supposedly settled after Zhizhi, was established under Emperor Wu of Han in 104 BCE, sixty-eight years before the Battle of Zhizhi. The county already existed, with that name, when Crassus crossed the Euphrates [4]. A garrison settled there in 36 BCE could not have given the place its name, because the place was already named.
Dubs was aware that Liqian predated Zhizhi, and he argued that the resettled legionaries were placed in an existing county whose name happened to fit. The argument is possible. It is also a long way from his original suggestion that “Liqian” was a Han transliteration of “legio,” because the Han bureaucracy did not generally rename a sixty-year-old administrative unit to honor newly arrived prisoners of war. The simpler reading, advanced by Yang Gongle and others, is that Liqian was an early-Han name for the Hellenistic kingdoms of Bactria and Ferghana, and that the older Chinese sources used “Lijian” or “Lixuan” for Greek-speaking populations along the Silk Road generally, only later narrowing to mean Rome [4].
What the DNA Did and Did Not Show
Two waves of genetic testing, both run on Liqian villagers, have been read by tabloid headlines as confirming the legend. The actual results were considerably more sober.
The first study, by Ruixia Zhou and colleagues, appeared in the Journal of Human Genetics in 2007 [5]. It analyzed Y-chromosome short-tandem-repeat and single-nucleotide-polymorphism markers in 227 male individuals from four populations in the region. Seventy-seven percent of Liqian Y chromosomes belonged to haplogroups restricted to East Asia. Principal-component analysis placed Liqian close to Han Chinese populations and far from Central Asian or Western Eurasian samples. The authors’ conclusion was deliberately deflating: their data did not support a Roman mercenary origin for the modern Liqian population.
The second wave, conducted by Lanzhou University in 2008 and reported widely in 2010, sampled eighty-seven villagers and found three with markers the team described as West Eurasian. A handful of villagers, including a man named Cai Junnian whom journalists nicknamed “Cai the Roman,” tested as carrying significant European-origin admixture. None of this identifies a Roman source specifically. Western-Eurasian admixture in Gansu is unsurprising on its own terms; the Tarim Basin to the west was home to the Indo-European-speaking Tocharians for nearly two millennia, and the Silk Road carried genetic exchange in both directions for centuries [6]. The DNA shows what one would expect of a population on a long-running East-West interface; it does not, on its own, point to Italy.
The Romance and the Discipline
There is a reason this hypothesis has been retold every decade since 1941, and it is not a perverse one. The romance lies in the symmetry. Two ancient empires that knew of each other only through Silk Road intermediaries, that traded silk and glass through five layers of middlemen, are made to almost touch through the bodies of a few hundred prisoners of war. It is the sort of historical encounter the human imagination wants to be true. Christopher Beckwith, in Empires of the Silk Road (Princeton, 2009), gives the chapter on this period the title “Between Roman and Chinese Legions” precisely because the cultural distance is the point: the two civilizations operated as distant gravitational masses, deflecting one another at long range while almost never meeting in the flesh [7].
The discipline of holding the question open lies in noticing what would have to be true for the romance to hold up, and how thin the textual ground is when one looks. One Han report. One ambiguous formation metaphor. One county name with at least two reasonable etymologies. One DNA dataset showing the genetic exchange one would expect of a frontier village without showing a specifically Roman fingerprint. None of these forbid Dubs’s reading. Each, taken on its own terms, supports a different and simpler reading more readily.
What survived from Carrhae, almost certainly, is what survives from most Roman defeats: a rumor in Plutarch, a few prisoners absorbed into Parthian society, a great deal of silence. What sits at the edge of the Gobi today is a village whose people are demonstrably Han with a small admixture consistent with a long Silk Road. The story between those two facts is a beautiful one. It is also, as far as the evidence will currently bear, a story.
Sources for Further Reading
- Britannica’s entry on the Battle of Carrhae for the Roman side of the disaster.
- The Wikipedia summary of Liqian, which gathers the textual, genetic, and historiographic critiques in one place with sources.
- Zhou et al.’s 2007 Y-chromosome study at the Journal of Human Genetics for the primary genetic data.
- For the wider context of Roman-Chinese near-contact, the Historical & Archaeological Mysteries hub at Eso Vitae assembles related cases on the long edges of the Silk Road.
