Half the French paintings made between 1840 and 1929 in two American museums are on a canvas bought off the shelf: both sides within a centimetre of a colourman's numbered stock size.
Paris colourmen sold primed canvas ready-stretched in sixty fixed sizes, numbered 0 to 120 in three shapes called figure, paysage and marine. Nobody recorded which painters bought them. The paintings did: their measurements sit in two museum databases, taken by different hands, in different units. Every figure below is computed in your browser from 5,975 paintings on canvas.
Sixty rectangles
Each painting is a dot at its long side and short side. Each little square is one of the colourman's sixty stock sizes, drawn one centimetre wide each way: the tolerance. A dot inside a square is on the grid. Hover a dot to see the painting.
figurepaysagemarinepainting on the gridpainting off it
Stretch the catalogue
Sixty squares cover a fair share of the plausible sizes, so some paintings land on the grid by chance. To see how many, multiply every stock size by a constant and count again. A real catalogue should fit at exactly one scale and nowhere near it. This is the control: if a decoy catalogue four per cent or more away ever catches half as many paintings as the real one, the headline above is withdrawn.
Share on the grid, as the catalogue is shrunk or stretched
French paintings, 1840 to 1929, each museum on its own line. The faint lines are American paintings of the same years tried against the same French catalogue.
Metropolitan MuseumArt Institute of ChicagoAmerican paintings, same years
The wrong catalogue
A second control, from the other side of the Atlantic. In about 1900 F. W. Devoe & Co. of New York listed forty-five stock stretched canvases in inches (16 by 20, 20 by 24, 25 by 30). If the method finds grids that are not there, French paintings will fit the American list too. They do not: they fit it no better than a decoy. American paintings fit their own list, and a sixth of them fit the Paris list as well.
When the grid arrived
French paintings by twenty-year span of their start date. The grey ticks are the decoy level for the same paintings. Spans with fewer than twelve paintings in a museum are left out, and the line breaks there. The grid does not start with the Impressionists: at the Met it is already well above chance in the early 1700s, on a few dozen paintings.
Figure, for everything
The three shapes carry subject names: figure, landscape, seascape. The folklore of the trade is that they were cut for those subjects. Among the French paintings that sit on the grid, here is which shape each subject was painted on, using each museum's own subject tags. Both museums agree on the big thing: the near-square figure held most of every subject, the sea included, and marine was the rarest shape for every subject. They disagree on the small one, and the numbers there are too few to settle it.
Metropolitan Museum
Art Institute of Chicago
Who bought them
Every artist with at least six canvases dated 1840 to 1929 in a museum, whatever their nationality. The two museums spell names differently and are listed apart. Pierre Bonnard, who pinned unstretched canvas to his wall and cut each picture out when it was finished, is the painter the method should miss, and it does. The Americans split by where they worked: Mary Cassatt, who lived in Paris, is mostly on the grid; John Frederick Kensett, Winslow Homer and George Inness, who did not, almost never are.
Metropolitan Museum
Art Institute of Chicago
Methods
Sources
The Metropolitan Museum of Art, Open Access CSV (metmuseum/openaccess, MetObjects.csv, 318 MB through Git LFS), read . Object data CC0. Every object classed Paintings, or, in the American Wing, named Painting, whose medium mentions canvas.
Art Institute of Chicago, public API (api.artic.edu), every artwork typed Painting, read . Data CC0 (descriptions, not used here, are CC BY). Medium mentions canvas.
The French stock sizes: the table of the sixty figure, paysage and marine sizes on English Wikipedia, French standard sizes for oil paintings (CC BY-SA 4.0), which cites Beatrix Haaf, Zeitschrift für Kunsttechnologie und Konservierung 1 (1987). It is a modern chart; see the failed half below.
The American stock sizes: F. W. Devoe & Co., Priced catalogue of artists' materials, New York and Chicago, about 1900, pages 16 and 17, "Single Prime, Smooth, or Roman Canvas on Stretchers", scanned by Columbia University Libraries for the Internet Archive (public domain), read from its OCR text.
What was measured, and how
Size. The first height and width in centimetres in each museum's dimension string that is not labelled as the frame. The Met measures in inches to the nearest eighth and gives centimetres in brackets; Chicago gives centimetres first. Paintings whose dimension string mentions an added strip, an enlargement or a sight size () are kept on the scatter and left out of every rate.
On the grid. Long and short side both within 1.0 cm of the same stock size (the larger of the two differences is at most 1 cm). Orientation is ignored: a stock canvas can hang either way.
Decoy catalogues. The sixty sizes multiplied by 0.90, 0.92, 0.94, 0.96, 1.04, 1.06, 1.08 and 1.10; the decoy level is the mean of those eight. The sweep chart runs every scale from 0.80 to 1.20.
Two measures, never pooled. The Met and the Art Institute catalogued their paintings independently, in different units and different formats. Every rate is computed for each museum separately and the two are shown side by side; nothing on this page averages them.
Nationality is the artist's, as each museum records it (first listed, before any comma). Subject comes from the museums' own tags: any sea, boat, ship, harbour or beach tag makes a seascape; otherwise a portrait tag a portrait; otherwise a landscape, river, mountain or forest tag a landscape.
Instrument findings
The Met's American Wing has no classification. All 18,532 of its objects leave the Classification column blank, so the obvious filter, Classification == "Paintings", silently drops its 1,006 paintings on canvas. Before the fix, the American control rested on 8 paintings from 1840 to 1899 instead of .
Chicago's structured sizes round down to the centimetre. The API's dimensions_detail field gives La Grande Jatte as 207 × 308 when the label says 207.5 × 308.1. That is a shift of up to a centimetre, the whole tolerance, and it does not fail safe: the stock sizes are whole centimetres, so cutting the decimals snaps a canvas measured at 73.4 onto 73 exactly. Measured from the truncated field, Chicago's French rate would be instead of , a finding flattered by the database rather than the paintings. This page parses the text label instead.
Chicago's search stops at 1,000 results. Asking for page 11 of a 100-per-page query returns HTTP 403, not an empty page. The fetch is split into date ranges that each hold fewer than a thousand paintings and stops if one does not.
Where the measurements fall. For French paintings within 2.5 cm of a stock size, the median overshoot is . A museum measures a canvas as it is now, stretched and sometimes relined, so a few millimetres over the nominal size is what one would expect. The peak sits on zero in both museums.
The failed half
No period French catalogue. The plan was to take the French sizes from a nineteenth-century colourman's price list rather than a modern chart. A full-text search of Gallica returned one candidate: the Menier Catalogue commercial ou prix courant général of 1860 (ark:/12148/bpt6k946633g, 690 pages), whose text contains figure, paysage and marine. Its text view sits behind an ALTCHA proof-of-work page, and after the first ALTO page gallica.bnf.fr refused every connection from this address for the rest of that session. When it answered again, its own in-document search showed the hit was false. Menier was a wholesale druggist; its paysages and marine are painted window blinds, and it lists no artists' canvas at all. The real candidates, Sennelier's catalogues of 1896, 1902 and 1912, are held by the Bibliothèque Forney and found through the same search, but its viewer is a JavaScript application with no image endpoint this build could reach. The other places searched (archive.org full text, HathiTrust behind a Cloudflare check, the painting manuals of Bouvier, 1827, and Arsenne, 1833) gave no table. Bouvier mentions canvases sold ready-stretched but advises buying a roll, to be free to vary the proportions.
What stands in for it. Only indirect evidence that the modern chart matches the old sizes: the measurements of the paintings themselves peak within a few millimetres of it in both museums, and a chart off by even three per cent would lose most of them (see the sweep). That is evidence about the paintings, not about any catalogue, and it is weaker than a period price list would have been.
What this does not say
It does not say any single painting was bought ready-made. Stretchers were sold in the stock sizes on their own, and a painter could tack his own canvas to one. A dot on the grid says the support is a stock size, not who stretched it.
It is not a sample of French painting. It is what two American museums chose to collect, heavy on the Impressionists and on easel pictures. Large Salon machines and decorative commissions are rarer here than they were in Paris.
Nationality is a proxy for where the canvas was bought. A Spaniard or an American working in Paris bought Paris canvas; the Spanish rate is high for that reason.
A canvas relined, trimmed or with its tacking margins opened out may have left the grid after it left the shop. That can only lower the rates.
The subject tags are the museums' and are uneven. The seascape test rests on a few dozen paintings in each museum.
Code
MIT. The fetch and shaping scripts are in the repository's src/, and REBUILD.md lists the expected values.