The Queen Without a Face: Can AI Really Identify Anne Boleyn in a 500-Year-Old Sketch?

Nearly 490 years after her execution at the Tower of London, Anne Boleyn remains one of history's most visually elusive figures. Henry VIII reportedly ordered the destruction of her portraits after her death in 1536, and no image created during her lifetime has ever been definitively authenticated [1]. Into this void, a team of researchers has now aimed the tools of 21st-century facial recognition — and the result has ignited a fierce argument between computer scientists and art historians about what technology can and cannot tell us about the dead.

The Study: AI Meets the Holbein Collection

In March 2026, historian Karen L. Davies, computer scientist Professor Hassan Ugail of the University of Bradford, and David G. Stork of Stanford University published a study in npj Heritage Science, a peer-reviewed journal in the Nature portfolio [2]. Their paper reassesses two drawings by Hans Holbein the Younger held in the Royal Collection Trust.

For over two centuries, a Holbein sketch catalogued as RCIN 912189 has been labelled as Anne Boleyn, based on an unverified 18th-century inscription [2]. The researchers argue this identification is wrong. Using a deep learning model based on AdaFace embeddings — an algorithm presented at CVPR 2022 that adapts its recognition thresholds based on image quality — they compared facial features across the Holbein collection, cross-referencing them with authenticated portraits of Anne Boleyn's daughter Elizabeth I and her cousins [3][4].

Their conclusion: a different, lesser-known drawing — RCIN 912190, catalogued as an "Unidentified Woman" — shows stronger facial similarity to Elizabeth I and aligns more closely with contemporary written descriptions of Anne Boleyn, including her documented dark hair, slender build, and the "little neck" she reportedly referenced before her execution [2][5].

The team calls their approach a "Working Likeness Methodology" that restricts biometric comparison to Holbein's life-drawn sketches rather than later copies or posthumous portraits [2]. Without an agreed-upon reference portrait, they constructed a visual family tree, comparing facial geometries among known Boleyn relatives to triangulate which sketch most plausibly depicts Anne.

The Critic: "A Load of Old Phooey"

Art historian Dr. Bendor Grosvenor has rejected the findings in blunt terms. "They're trying to take a 21st-century technology, which is facial recognition and AI, and assume that it can just be put immediately back onto drawings and artworks made 500 years earlier in the 16th century," he told ITV News. "The methodology here is so profoundly misplaced that it almost is impossible for it to get off the ground in the first place" [6].

Grosvenor's objections are specific and substantive. He argues that facial recognition algorithms are trained on photographs — images that mechanically capture light reflected from three-dimensional faces. A Holbein sketch, by contrast, is an artistic interpretation filtered through the conventions, tools, and subjective choices of a 16th-century draughtsman [7]. Variations in line weight, shading technique, paper condition, and the artist's own stylistic tendencies introduce distortions that no algorithm trained on modern photography can reliably parse.

He further contends that the traditionally labelled sketch was identified by someone who knew Boleyn personally. The blonde hair visible in the drawing, he argues, results from a darker topcoat having rubbed off over centuries, and the subject's informal dress indicates high status consistent with a queen [6][7]. Grosvenor has also suggested that many of his art-historian colleagues privately agree but are reluctant to publicly challenge a peer-reviewed paper backed by computational methods [6].

The researchers, for their part, note that the inscriptions on the Holbein sketches were not made by Holbein himself but are attributed to Sir John Cheke, tutor to Edward VI — and Cheke was demonstrably incorrect in several of his other identifications [8]. This weakens the provenance chain that Grosvenor relies on.

The Remains That Aren't There

Any discussion of Anne Boleyn's "true face" must confront a fundamental problem: there is no verified biological evidence to anchor any depiction.

After her execution on May 19, 1536, Anne's head and body were placed in an elm chest originally used to store bow-staves and buried without ceremony beneath the chancel pavement of the Chapel of St. Peter ad Vincula at the Tower of London [9]. During renovations in 1876–1877, skeletal remains were exhumed. Dr. Frederick J. Mouat identified one set as belonging to "a female of between twenty-five and thirty years of age, of a delicate frame of body" with "particularly small" cervical vertebrae — which he interpreted as consistent with accounts of the queen's slender neck [9][10].

The bones were reinterred. No cast of the skull was taken. No DNA analysis has ever been performed [10][11]. As the Anne Boleyn Files, a leading scholarly resource, states plainly: "Unless DNA testing is performed on the remains, we'll probably never know" whether the bones are hers [10]. There is also no evidence that a death mask was made — Henry VIII's campaign to erase Anne's memory makes such preservation unlikely [11].

This means the 2026 study is not, strictly speaking, a facial reconstruction at all. Unlike the 2013 reconstruction of Richard III — which began with a verified skeleton found beneath a Leicester car park, confirmed through mitochondrial DNA matching with living descendants, and then built a face using tissue-depth markers applied to a CT-scanned skull — the Davies-Ugail-Stork study operates entirely within the domain of art attribution [12][13]. It asks which drawing most plausibly depicts Anne, not what Anne's face actually looked like in three dimensions.

How Accurate Is Forensic Facial Reconstruction?

Even when researchers do have a skull to work from, the science of facial reconstruction carries significant uncertainty. Accuracy studies report variable results, with the technique described as "easily the most subjective — as well as one of the most controversial — techniques in the field of forensic anthropology" [14].

Quantitative studies have found that the majority of reconstructed facial surfaces show less than 2.5 mm of error from the actual face, with 90% of male reconstructions and 75% of female reconstructions falling within 5 mm of accuracy. Recognition rates in blind tests — where volunteers are asked to pick the real face from a lineup — hover above 70%, compared to a 20% chance baseline [14][15]. But error concentrates at the ears and nasal tip, and soft-tissue features like lip shape, skin texture, and eye expression remain heavily dependent on the practitioner's artistic judgment [14].

The growing body of academic work in this field reflects both the opportunities and the controversies. According to OpenAlex, over 9,700 papers on forensic facial reconstruction have been published, with output rising from 167 papers in 2011 to a peak of 1,397 in 2025 [16].

Source: OpenAlex

Data as of Jan 1, 2026CSV

The parallel growth in research applying facial recognition to art history — over 44,500 papers since 2011, peaking at 6,248 in 2024 — suggests an expanding but still-young field where methodological standards remain in flux [16].

Source: OpenAlex

Data as of Jan 1, 2026CSV

A Long History of Trying to See Anne

The 2026 study is the latest in a string of attempts to pin down Anne Boleyn's appearance, each building on uncertain foundations.

The only confirmed contemporary image of Anne is a small lead medal from 1534, now held by the British Museum. It is severely deteriorated: the nose has been flattened, and one side of the face has been rubbed clear of defining features [7][17]. Everything else — the National Portrait Gallery painting, the Hever Castle portrait, the so-called "B" pattern portraits showing a woman wearing a pendant with the letter B — dates from decades or centuries after her death and was shaped by the political needs of the era in which it was produced [8][18].

In 2011, special effects artist Emily Pooley created a waxwork of Anne for Hever Castle, her childhood home. But Pooley did not work from skeletal remains; she found a living model who resembled a Holbein sketch, photographed her from 360 degrees, and sculpted from there [11]. The result is a likeness of a modern woman filtered through an artistic interpretation of an artistic interpretation.

More recently, scholars Elizabeth LaVasse and Richard Masefield traced the provenance of the "Lumley Portrait," a full-length painting possibly commissioned by Anne's father Thomas Boleyn during her lifetime but damaged by fire during Charles II's reign and last documented in 1926 [17]. If ever found, it could provide the closest thing to a verified reference image — but it remains lost.

Source: Various academic sources

Data as of May 6, 2026CSV

The contrast with Richard III is instructive. That reconstruction succeeded because the physical chain of evidence held: a skeleton was found, DNA-matched, CT-scanned, and then reconstructed — and the result bore a recognizable resemblance to surviving portraits, minus the propagandistic distortions added by Tudor-era artists who made Richard look villainous to please Henry VII [12][13]. For Anne Boleyn, no equivalent chain exists.

Peer Review, Funding, and Institutional Incentives

The Davies-Ugail-Stork study was published in npj Heritage Science, a Springer Nature journal with a standard peer-review process [2]. The involvement of Hassan Ugail, whose Centre for Visual Computing at Bradford has published extensively on facial recognition, and David Stork, known for computational analysis of art, lends the work credibility within the computer-science community [3][4].

But peer review in an interdisciplinary study raises a structural question: were the reviewers experts in Tudor art history, in computational facial recognition, or in both? Grosvenor's complaint — that "subject matter experts" in art history should drive the analysis, not software engineers applying algorithms designed for a different purpose — points to a gap that peer review across disciplinary boundaries does not always close [7].

The framing of the study's press coverage also warrants scrutiny. GB News headlined its story "Science breakthrough: Anne Boleyn's real face may finally be revealed after 'hiding in plain sight' for centuries" [19]. ITV asked, "Is this the real face of Anne Boleyn?" [6]. These framings imply a level of certainty the paper itself does not claim: the study proposes a reassignment of attribution between two Holbein sketches, not a definitive identification of Anne Boleyn's face.

Anne Boleyn occupies a unique position in the commercial ecosystem of British history. She has been portrayed on screen by Natalie Dormer in The Tudors (2007–2010), Natalie Portman in The Other Boleyn Girl (2008), and Jodie Turner-Smith in Channel 5's Anne Boleyn (2021) [20]. Philippa Gregory's Tudor novels, beginning with The Other Boleyn Girl, have spawned at least four television adaptations [20][21]. Google searches for Anne Boleyn spiked dramatically during The Tudors' original run [20]. This sustained public fascination creates an audience — and therefore a commercial incentive — for any claim to have discovered her "true" appearance.

This does not mean the researchers acted in bad faith. But the gap between what the paper actually argues (that one Holbein sketch is more plausibly Anne than another) and how it was packaged for public consumption (as revealing her "real face") is a pattern familiar from other historically themed research. Institutions benefit from media coverage; media benefits from dramatic headlines; and the public absorbs a more definitive narrative than the evidence supports.

The Unfalsifiability Problem

The steelman case against this research — and against all similar efforts — is straightforward: without confirmed remains, any claim about Anne Boleyn's appearance is unfalsifiable.

To actually prove that RCIN 912190 depicts Anne Boleyn, one would need either a verified contemporary portrait (none exists), authenticated DNA-matched remains from which a forensic reconstruction could be compared to the drawing (the remains at the Tower have never been DNA-tested and their integrity after the 1876 exhumation is uncertain), or a documentary source contemporaneous with Holbein explicitly identifying the sitter (no such source has surfaced) [2][10][11].

The researchers acknowledge the absence of a definitive reference image — it is, after all, the premise of their study. Their methodology attempts to work around this gap by using familial resemblance as a proxy. But this introduces its own circularity: the "authenticated" portraits of Elizabeth I and the Boleyn cousins are themselves subject to the same art-historical uncertainties about accuracy, artistic convention, and political purpose [8][18].

What remains after stripping away the headlines is a carefully argued but inherently limited proposition: given the available Holbein sketches, the pattern of familial features, and the textual descriptions from Anne's contemporaries, one drawing fits better than the other. That is a meaningful contribution to art-historical scholarship. Whether it tells us what Anne Boleyn looked like is a different question — and one that, for now, cannot be answered.

What Would It Take?

The path to a more definitive answer is clear in principle, if politically difficult in practice. DNA testing of the remains at the Chapel of St. Peter ad Vincula could confirm or exclude their identification as Anne Boleyn. If confirmed, a CT scan and full forensic reconstruction — following the Richard III model — could produce a science-based facial approximation [9][12].

But the Tower of London is a royal peculiar under the jurisdiction of the Crown, and past requests to test the remains have been declined. The bones were reinterred with Victorian-era ceremony, and disturbing them again would require navigating layers of institutional, ecclesiastical, and public sentiment [9][10].

Until that happens, every claim about Anne Boleyn's face — whether generated by algorithm, artist, or novelist — operates in the same epistemic space: informed speculation about a woman whose husband made sure the world would never quite be able to see her.