X-Radiography, Pentimenti and Reused Supports

What the plate actually records

An X-radiograph of a painting is a map of how much radiation each part of the object absorbed. Heavy elements absorb most, and in European easel painting the heavy element that matters is lead: lead white in the ground, lead white in the light passages, lead-tin yellow, lead-bearing driers. Iron earths, organic lakes and carbon black are close to transparent. What comes back is therefore not a view of the hidden layers but a density image in which the priming, the weave that the priming has filled, the tacks and nails, the panel joins and keys, the losses and the old fills all appear alongside the lead-rich brushwork.

That is why radiography is a structural instrument first and a pictorial one second. The standard method of counting the threads of a canvas uses a radiograph precisely because the lead-bearing priming makes the individual threads visible. It is also why radiography has always been used in combination: Joyce Plesters's examination of Rubens's Samson and Delilah for the National Gallery Technical Bulletin in 1983 determined the painting procedure and the paint layer structure through X-radiography together with microscopic analysis of paint cross sections, not through the plate alone.

The technique is old in forensic use. Radiographic examination was already part of the evidence in the Otto Wacker van Gogh forgery trial in Berlin in the early 1930s. Forgers have therefore had roughly ninety years to plan around it, which is a fact that belongs in the reading of any radiograph of a picture with a thin provenance.

The support testifies before the image does

Until recently, distinguishing one canvas from another meant counting threads: a pointer under magnification, four to fifteen sample patches, fractions estimated by eye, and the results averaged into a figure such as 12.5 by 17.2 threads per centimetre. C. Richard Johnson Jr. at Cornell and Don H. Johnson at Rice replaced that with algorithms that generate weave density maps across an entire radiograph, plus weave angle maps that graph deviations from the rectilinear.

The Vermeer results are a lesson in how easily thread counts mislead. Twenty-nine of the thirty-six generally accepted paintings have been mapped. Four gave averages so close as to invite conclusions: The Milkmaid at 14.4 by 14.7 threads per centimetre, Woman in Blue Reading a Letter at 14.6 by 14.7, Girl with a Pearl Earring at 14.2 by 14.1, and the Metropolitan's Study of a Young Woman at 14.3 by 14.8. The last two are tronies of almost identical size, 44.5 by 39 centimetres and 44.5 by 40 centimetres, which had encouraged the argument that they were painted together or conceived as a pair. Computer analysis showed that all four canvases come from four different bolts, and that none of them matches any of the other twenty-five.

Where a real match exists it does argumentative work. Libby Sheldon and Nicola Costaras proposed in 2006, from the radiographs, that Young Woman Seated at a Virginal and The Lacemaker were on the same coarse fabric at 12 by 12 threads per centimetre. The automated counts refine that to 12 by 12 against 11.9 by 12.4, a looser match than the Milkmaid pairing, but the weave maps confirm a common bolt, and that confirmation supports one of the technical arguments advanced for the attribution of the Virginal picture to Vermeer.

Weave angle maps carry the other half of the evidence. They show primary and secondary cusping, the scalloped distortion produced when a canvas is stretched and primed, and weft snakes, wandering weft lines produced on a hand loom, which appear in about 40 per cent of the Vermeer canvases surveyed. Cusping along an edge indicates that the canvas was primed while stretched and that the edge has not been cut away; its absence raises the question of what happened to the missing margin. The panel equivalent is just as blunt: the National Gallery of Art's Girl with a Flute retains untrimmed bevelled edges on the reverse, which is how the examiners knew it had not been cut down.

Changes of mind evidence invention, and stop short of a name

The classic attributional use of radiography is the pentimento. A copyist works from a finished surface and has no reason to change his mind about a composition that already exists; a painter inventing one does. Reworkings, abandoned contours and entire buried compositions are therefore evidence that a picture was being made rather than reproduced.

The strongest recent examples come from macro X-ray fluorescence scanning, which maps elements across a whole painting and produces a colour map of the hidden layers that radiography can only hint at. In 2011 the Rijksmuseum announced that Goya's Portrait of Don Ramon Satue of 1823 concealed an unfinished formal portrait of a man in uniform. The decorations are those of the highest rank of the Royal Order founded by Joseph Bonaparte as king of Spain, which dates the hidden portrait to between 1809 and 1813; only fifteen generals plus Joseph himself were entitled to wear them, and the sitter has not been identified. The technique was developed by the University of Antwerp and Delft University of Technology, and had been tried two years earlier on Van Gogh's Patch of Grass of 1887, under which it found the head of a peasant woman.

Now the corrective, and it is decisive. The National Gallery of Art's Girl with a Flute shows revisions to the arm and the flute, and paint defects severe enough that the artist scraped the surface level before applying the final layers. It has changes of mind, and in 2022 the Gallery concluded that it is not by Vermeer but by an artist working in his studio. Conversely, the mechanical underdrawing of the Metropolitan's Christ Appearing to His Mother, laid in without changes and probably from a pattern or tracing, was one of the observations that demoted it from Rogier van der Weyden. Pentimenti bear on whether a design was invented on this support. They do not name the hand that invented it.

Reuse cuts both ways

A reused support is often presented as a mark of authenticity, on the reasoning that a forger would not sacrifice an old painting. The two most consequential forgers of the last century both did exactly that.

The radiograph of Christ and His Disciples at Emmaus shows traces of paint from the original picture van Meegeren scraped off in order to obtain a seventeenth-century canvas. The technical study of Beltracchi's work found the same manoeuvre repeatedly: grounds sanded before painting, canvas threads left exposed beneath the paint, and tacking margins lightly patinated to look dirty, all consistent with an earlier image having been removed from an old fabric.

Beltracchi went further, and his own account is the most useful sentence in the literature for anyone who reads radiographs. He could not remove the original painting from the canvas of Rotes Bild mit Pferden completely, so part of his new composition was dictated to him: a diagonal rising from left to right, in the old picture a path with buildings and a large tree in the middle ground, which he integrated into his design so that those elements would not appear as anomalies under X-ray examination. The radiograph does show parts of an old, discarded composition, and it shows them looking like an artist's abandoned idea.

The evidential rule that follows is narrow and useful. A reused support is a fact requiring explanation, not a certificate. It convicts only when the underlying image itself carries later stylistic features, which is the one thing a forger cannot easily control if he cannot scrape the canvas clean.

What radiography cannot reach

It cannot date anything. Nothing in a density image carries a year, and the technique's contribution to chronology is always indirect, through the support.

It cannot identify a hand. Handling is legible in the radiograph as a pattern of lead-rich strokes, and specialists do read it, but the National Gallery of Art's demotion of Girl with a Flute rested on the comparison of paint handling at high magnification and on the sequence of preparatory layers, not on any single image. The full apparatus there ran to X-ray fluorescence element mapping, reflectance imaging spectroscopy, multispectral infrared reflectography, fibre-optic reflectance spectroscopy, polarised light microscopy of cross sections, and reflectance transformation imaging with specular enhancement to raise the surface texture. Radiography was one channel among many.

And it is easily blinded. Lead white in overpaint and in old fills reads exactly like lead white in original paint, so a heavily restored picture returns a muddled plate. Vermilion and lead-tin yellow are also strong absorbers and can drown the passages beneath them. A canvas on a lead-containing ground can obscure its own weave. Cradles, wax-resin linings, added strips and thick putty fills all add density that has nothing to do with the artist. A radiograph should therefore always be read after the treatment history, not before it, because half of what appears on the plate may have been added by conservators.