What Technical Art History Can and Cannot Prove

Every technique is stronger at exclusion than at confirmation

The asymmetry is the first thing to understand and the last thing that survives translation into a press release. Jane Kallir stated it in a single sentence: the identification of pigments unknown during an artist's lifetime may be used to expose a forgery, but the presence of historically appropriate materials does not necessarily prove authenticity. A negative finding can be decisive; a positive one establishes consistency.

Matthijs Ilsink, writing for the Bosch Research and Conservation Project after years of examining one artist's corpus with every available instrument, applied the same logic to the most respected of the dating techniques. Dendrochronology, he wrote, can only reject paintings, and then only if they are on oak or another wood for which there are chronologies. That is a scientist describing the limits of his own strongest tool.

The consequence for reading any technical report is procedural. Ask what the report excludes. A document that lists consistent findings and stops has narrowed the field without closing it, and a laboratory has no mechanism for the final step of naming a person. Ilsink's summary of what all the instrumentation actually established remains the honest one: connoisseurship has played a part in every single attribution to Bosch as matters stand today.

Dendrochronology dates the tree, and the inference does the rest

Tree-ring dating measures the growth rings in a wooden panel and matches them against regional chronologies. Peter Klein, the standard practitioner, describes the output precisely: the method allows a terminus post quem for a panel painting, the earliest date at which the painting could have been executed, by establishing a plausible felling date for the tree from which the panel was cut.

Every step after the last measured ring is an assumption with a published range. Klein states that at least nine sapwood rings must be accounted for to determine the earliest possible felling date for an oak. His sample of oaks from northern Poland produced a median of fifteen sapwood rings, with a minimum of nine and a maximum of thirty-six, and half the trees falling between thirteen and nineteen. For wood originating in Germany or the Netherlands the median was seventeen, with half between thirteen and twenty-three and a minimum of seven. Seasoning adds more: for most oak panels of the sixteenth and seventeenth centuries the seasoning time has been determined at approximately two to eight years, while fifteenth-century studies indicate an average of ten to fifteen years.

Worked forward, the method behaves well. Klein's analysis of Rembrandt's Two Old Men Disputing gave a last ring of 1603, an earliest felling of 1612 and an earliest painting date of 1620 onward, which sits comfortably with the date of 1628 on the panel itself. Worked backwards, it settles less than people expect: Klein notes that where a felling date looks significantly earlier than a proposed date of execution, the wood may simply have been stored for a long time before being painted.

The technique fails silently, and the failures are instructive

Two of the panels in the same National Gallery of Victoria study could not be dated at all, and the reasons are worth more attention than the successes. For a Virgin and Child by Simon Marmion, dating by dendrochronological analysis was not possible: the panel had been thinned and glued to a nineteenth-century board, destroying the evidence. For a Madonna and the Child by a follower of van Eyck, no dendrochronological solution was possible because the ring sequence matched no existing master chronology.

Both failures are structural rather than accidental. Historic conservation practice routinely thinned panels, and the master chronologies cover particular regions and particular periods. A picture on a wood for which no chronology exists, or a panel that a nineteenth-century restorer planed down, produces no result at all, and the absence of a result is frequently reported as though the technique had not been tried.

This is the pattern across the discipline. The instruments do not fail loudly. They return nothing, or they return something consistent with several answers, and the report is then read as neutral when it should be read as uninformative. A commissioning party that expected a verdict receives a range, and the range is where all the argument lives.

Infrared reflectography, and one claim that is weaker than it sounds

Infrared radiation above roughly 760 nanometres passes through varnish and some pigments and reaches the carbon-based underdrawing beneath. Infrared films were sensitive to about 1100 nanometres; modern cameras reach around 2400. The Museo Nacional Thyssen-Bornemisza describes what the resulting images show: drawings, corrections or pentimenti, reuse of canvases, hidden inscriptions, the marks artists used as a reference, and possible corrections or alterations made to the composition by the artist.

Its evidential value for identifying a hand is moderate, because underdrawing habits are shared within a workshop. Its value in an attribution campaign is real nonetheless. The Bosch Research and Conservation Project used infrared reflectography together with ultra-high-resolution macro photography and microscopic study to conclude in 2015 that Christ Carrying the Cross in Ghent and The Seven Deadly Sins in Madrid were likely made by imitators rather than by Bosch, and to attribute a previously unknown drawing, Infernal Landscape, to him.

Connoisseurs also use infrared for a second, sharper distinction: a drawing full of changes, hesitations and abandoned contours indicates a composition being worked out, while a confident continuous drawing that exactly matches the paint above suggests a design transferred from something that already existed. That inference is widely applied and is a reasonable reading of what the images show, but it is a working connoisseurial criterion rather than a published institutional rule, and it should be presented as the argument it is.

Pigment analysis is the backbone, and restoration contaminates it

Sampling identifies the materials present, which can be checked against the date those materials became available. Titanium dioxide white was discovered in 1821, entered commercial production in 1916 at the Titanium Pigment Corporation in Niagara Falls and the Titan Company in Norway, and was not introduced as an oil colour suitable for artistic purposes until 1921. Prussian blue is the oldest modern synthetic pigment, in use since its discovery in 1704. Zinc white was known as a mineral since antiquity but was not used as a pigment until the eighteenth century. Each supplies a hard floor beneath which a claimed date cannot go.

The complication is that a painting four hundred years old has usually been cleaned, retouched, relined and revarnished, sometimes repeatedly, and the analytical practice makes the problem worse rather than better. The University of Delaware's conservation resources describe cross-sections as giving information on paint stratigraphy, pigments, and the presence of unoriginal restoration layers, and then state where samples come from: paint samples are generally collected along the edges of paintings, or near extant losses or cracks in the paint and ground layers. Those are exactly the locations where a restorer worked.

Modern material in a surface sample may therefore mean a forgery, or it may mean a competent conservator in 1963. Sampling has to be located rather than merely taken, and the results have to be read against the conservation record. The Delaware guidance adds a constraint that rarely reaches the market: ethical factors are always considered when choosing a sampling location, and occasionally sampling is judged inappropriate for a given artwork altogether.

X-radiography shows structure, and reuse proves less than it seems

X-ray transmission through a material is inversely proportional to the atomic weight of its elements, so radiography images the heavy pigments: lead white above all, with lead-tin-yellow and vermilion. What it therefore reveals is structure. Bellini's Feast of the Gods showed under X-ray that the background behind the foreground figures originally consisted of single tree trunks, that Bellini had altered the foreground, and that the left part of the background was overpainted after his death by Titian.

The technique also has a resolution ceiling that is rarely acknowledged. Rembrandt's An Old Man in Military Costume conceals an underlying portrait of a young man, painted with the support rotated by 180 degrees. Conventional radiography revealed that something was there, but the quality of the image was not sufficient to discern any details; it took X-ray fluorescence and neutron activation autoradiography to produce usable detail in the face and cloak of the hidden figure.

One inference commonly drawn from these images deserves qualifying. It is often said that an abandoned composition beneath the surface proves a working artist rather than a copyist, since a forger has no reason to paint a picture he intends to hide. The Phillips Collection, describing the earlier portrait beneath Picasso's The Blue Room, gives a different and more mundane explanation: reusing canvases was a common practice for Picasso at the time, since he was quite poor. Reuse is evidence of economy at least as much as of invention, and the stronger version of the claim is not supported by the institutions that publish these images.