Five bright points swirl at the center of Vincent van Gogh’s “The Starry Night.” A team of South Korean researchers ran those five points through a shape-matching algorithm called Procrustes analysis and got a match: the Hyades, a cluster of hundreds of stars in Taurus, five of which are visible to the naked eye. The algorithm found that match closer than any alternative constellation it tested, including Aries, the constellation two earlier studies had proposed. The result, published October 22 in Scientific Reports, pins the sky in the 1889 painting to the predawn hours of July 22 of that year — a month after the date art historians have assigned the canvas since Vincent wrote to his brother Theo about a “new study of a starry sky” around June 18.
Procrustes analysis is a standard tool in shape statistics, not an astronomy-specific one. It strips a set of points of its position, its scale and its rotation, leaving only the relative geometry — the skeleton, in the researchers’ word — so that a cluster drawn too big, tilted, or shifted across the canvas can still be compared fairly to where the real stars stood that night. Lead author Elijah Jeong-hyun Kim, an astronomer and engineer who is CEO of the Korean space-science firm SLLAB and a doctoral candidate at Kyung Hee University, told CNN the method “removes differences in position, size and rotation and then compares only the ‘shape’ of two sets of points,” as CNN reported. Run against a simulated sky over the hills southwest of the Saint-Paul-de-Mausole asylum near Saint-Rémy-de-Provence, where van Gogh was a patient, the five-star skeleton of the painting’s central cluster lined up with the Hyades’ V-shape rising alongside Venus — and, per ScienceAlert’s account of the paper, every other constellation pairing the team tried produced a worse fit.
| Criterion | June 18, 1889 (traditional) | July 22, 1889 (proposed) |
|---|---|---|
| Moon phase | Waning gibbous, more than half illuminated | Waning crescent, matching the slender shape painted |
| Hyades/Venus position | Not aligned with painted arrangement | Hyades rising alongside Venus, matching painted layout |
| Basis for date | Letter to Theo mentioning a new starry sky | Procrustes shape match plus moon and planet positions |
| Van Gogh's seizure | Assumed to begin mid-July, possibly blocking observation | Reassessed letters and weather data suggest late-July onset |
Source: Kim et al., Scientific Reports (2026); ScienceAlert; CNN
The moon is the detail that does the most work against the June date. Around June 18, 1889, the moon was waning gibbous — more than half lit — while the canvas shows a slim crescent. Around July 22, the moon had thinned to a crescent close to what van Gogh painted, and the Hyades cluster happened to be rising in the east with Venus at the same hour, matching the painting’s layout: moon high in the top right corner, Venus center-left, the star cluster between them. “Around June 19 is close to the summer solstice, when no impressive stars rise in the pre-dawn sky,” Kim told CNN. “Around July 22, by contrast, the Hyades rose in the east together with Venus, and the moon’s shape was close to the crescent in the painting.”
This is not the first astronomical pass at the painting, and it is worth saying what came before it mattered to how the new paper argues its case. Albert Boime proposed in 1984 that the canvas showed Aries and the dawn sky of June 19, 1889; Charles Whitney, writing in the journal Art History in 1986, also read Aries into the cluster but dated the scene to June 17–18 and called the sky “a celestial composite.” The two conclusions conflicted, and a separate 1986 Art Bulletin paper by Lauren Soth flagged the “precariousness” of reading astronomy off a painting at all — a caution the new authors inherit rather than resolve. Kim’s stated reason for revisiting the question, as he put it to reporters, was that Aries’ stars are faint enough that they would barely register to a painter at dawn, and that van Gogh’s letters mention watching Venus repeatedly but never mention Aries.
The study’s second move is to use a different van Gogh drawing, one depicting the same hillside and village, to fix where on the ground he was standing, which in turn fixes which patch of sky was over the horizon. That geographic pin is what lets the researchers run a specific-night simulation rather than a generic one. The same shape-matching method, applied to van Gogh’s 1888 painting “Café Terrace at Night,” turned up a correspondence with the Summer Triangle, the bright asterism formed by Vega, Deneb and Altair — a secondary result the paper treats as supporting evidence that van Gogh’s skies generally track real configurations rather than invented ones.
The seizure problem
The July date creates a biographical complication the researchers had to address rather than ignore: van Gogh is conventionally held to have suffered a seizure in mid-July 1889, an episode that would have made outdoor night painting difficult or impossible around the proposed date. The team re-read the original French and Dutch letters alongside contemporary weather records and found grounds to push the seizure later — into late July, and lasting, by their account, only a day or two, which would leave a window either side of July 22 for van Gogh to have observed the sky even if he could not have painted at the exact hour.
That revision is the part outside reviewers treat most carefully. Yongxiang Huang, an associate professor at Xiamen University’s Center for Marine Meteorology and Climate Change who was not involved in the study, called it “an engaging framework that combines astronomical modeling, historical meteorological data, and letter analysis,” while warning CNN that “astronomical alignments on a canvas must be interpreted with caution” because “Vincent van Gogh was a Post-Impressionist master, not an optical camera.” Huang is the author of a 2024 analysis arguing the painting’s swirls statistically resemble Kolmogorov turbulence in fluid dynamics — itself a claim about hidden physical regularity in a canvas usually read as pure expression, and one that drew the same category of skepticism this study now gets from him.
Ulrike Kuchner, an art-science researcher not involved in the paper, raised the plainer objection: van Gogh had artistic license, and might simply have preferred the look of a crescent moon to the gibbous one actually overhead in June. “Being an artist, Vincent likely mixed ‘imagination’ and ‘observation,’” she told CNN, calling the paper “exciting detective work” while noting that if the date moves, “the history books on his medical history need updating too, because the timing of his seizure, or his ability to paint afterwards, would have to be revised.” She also pointed out what the paper does not settle: whether the canvas is observation transcribed or fantasy built from fragments of memory remains, in her words, a question “never really addressed” by shape-matching five stars.
The study’s own limits are structural, not just a matter of outside caution. Nineteenth-century weather records predate standardized meteorological observation and carry less precision than modern instruments. The method cannot date when van Gogh started the canvas or whether he reworked it later; the researchers note the canvas has patches left unpainted around the stars, consistent with a plan sketched in advance and filled in afterward, possibly quickly. Kim told CNN the team’s own view, offered as opinion rather than a result of the analysis, is that van Gogh “conceived the idea in June and completed the painting after July 22” — and that the bare patches of canvas “fit with the possibility that the work was done in a hurry.”
What would settle this is more and better primary material: French weather station logs from the summer of 1889 with finer time resolution than what survives, and any letters van Gogh sent to other painters that mention the canvas directly, none of which the current paper had access to. Kim says he is looking for both. Until one surfaces, the claim rests on how well five points on a 92-centimeter canvas match the geometry of a star cluster 150 light-years away on one specific pre-dawn — a fit the authors call close, and that at least two outside readers call suggestive rather than settled.