Here is the single most useful thing to understand about the lab science behind a painting. It is very good at telling you what a work is not, and much weaker at telling you what it is. A pigment that did not exist yet cannot appear in a work supposedly painted before it, so when a lab finds titanium white in a canvas dated 1914, the case is closed: the painting is later, full stop. That is exclusion, and it is the real power of forensics. Proving the positive, that this specific Rothko came off Rothko's brush, is something the instruments cannot do on their own.

We treat this distinction as a piece of investing discipline, because confusing the two is how people lose money. A clean scientific report does not mean a work is authentic. It means the lab found nothing that disqualifies it, which is a different and weaker claim. This article walks through the actual methods, what each one detects, and the worked cases where the science caught a fake. It is the deep technical companion to our broader piece on how authentication works as a whole, which covers connoisseurship, provenance, and the archives. Here we stay in the lab.

What scientific methods are used to authenticate a painting?

The toolkit splits into two families: imaging, which looks through and across the surface without touching the chemistry, and materials analysis, which identifies the actual substances in the paint and the support. Labs run them roughly cheapest and least invasive first, because many of the chemical tests require lifting a microscopic sample, and you do not start cutting into a canvas before the imaging has told you where to look.

On the imaging side, four techniques do most of the work. X-radiography sends X-rays through the whole object and images density differences, which exposes hidden compositions, earlier sketches, and the construction underneath: nails, tacks, canvas weave, panel joins. Infrared reflectography reads in the near-infrared band, around 900 to 1700 nanometers, where many dark paints go transparent and the artist's underdrawing becomes visible, the charcoal or ink sketch a copyist often skips entirely. Ultraviolet fluorescence lights up varnish and later restoration, so an "untouched" old master that is actually 40% repaint gives itself away. And raking light, the simplest tool of all, just throws light across the surface at a shallow angle to read brushwork, impasto, and the craquelure (the network of age cracks).

On the materials side, X-ray fluorescence, or XRF, is the workhorse. It excites the surface with X-rays and reads back element-specific signals, telling you the painting contains, say, titanium, zinc, lead, cobalt, or cadmium. Macro-XRF scans the whole surface pixel by pixel and maps where each element sits. Raman spectroscopy and FTIR go a level deeper and read the molecule, not just the element, which matters because copper shows up identically in azurite, malachite, and verdigris, three different pigments. Radiocarbon dating ages the organic support, the canvas fiber or the wood. Dendrochronology counts tree rings in a wooden panel to date the wood. None of these names the artist. Every one of them can rule a date in or out.

What can an X-ray or infrared scan reveal about a painting?

Imaging tells you about process and structure, and process is hard to fake. A real painting is a record of decisions: the artist moved a figure, painted over a first attempt, sketched the composition before committing. A copy, by definition, skips that struggle, because the copyist already knows where everything goes.

X-radiography catches this in two ways. It reveals pentimenti, the changes an artist made mid-work, which dense pigments like lead white render visible under the final surface. A complex, evolving composition underneath is consistent with original creation. It also reads the physical construction. If a panel is claimed to be 16th-century Netherlandish but the canvas weave, tacking edges, or stretcher type belong to the 19th century, the attribution has a problem before anyone discusses brushwork. In the Beltracchi case we come to below, X-ray showed an underlying composition that did not fit the supposed history.

Infrared reflectography goes after the underdrawing. Many genuine old masters drew their compositions first, and that drawing, invisible to the eye under the paint, shows up under infrared. A lively underdrawing with corrections suggests an artist working out a picture. A supposed master with no underdrawing where the workshop always used one, or with a mechanically transferred outline traced from a reproduction, raises a flag. This is precisely the kind of analysis used on the Knoedler forgeries, where high-resolution imaging exposed traced, hesitant signatures inconsistent with the originals.

Here is the honest limit. Neither method identifies a single pigment or produces a date. They show density and they show drawing. Plenty of genuine artists painted directly with no underdrawing, and plenty of forgers now study an artist's process and reproduce it. Imaging narrows the field and flags the obvious problems. It rarely closes a case by itself.

How does pigment analysis catch a forgery?

This is where the science gets decisive, and it works through a simple logic: chemistry has a calendar. Specific pigments entered the artist's palette at specific, documented moments, and a material cannot appear in a painting made before that material existed. XRF and Raman read what is actually in the paint, and a single anachronistic pigment in the original layer can take a seven-figure attribution to zero.

The dates worth committing to memory are these. Prussian blue was discovered around 1704 in Berlin and in common use by the 1720s, so it has no business in a genuine Renaissance painting. Cobalt blue arrived commercially in the early 19th century. Zinc white became a practical artist's white around the 1830s. Cadmium yellows came in the 1820s to 1840s. And titanium white, the most famous tripwire in the whole field, did not appear in commercial artist's oil paint until the early 1920s. One XRF study dates the incorporation of titanium into paint pigments to 1921. Detect titanium in an original paint layer and the work is 20th-century or later, no matter what the signature says.

The Beltracchi case is the textbook demonstration. Wolfgang Beltracchi forged works in the styles of early-20th-century masters and sold them for years; German courts convicted him in 2011 on 14 proven forgeries that had fetched roughly $45 million combined, with total estimated profits above $100 million. What undid him was a tube of paint. He hand-mixed his own pigments specifically to avoid modern materials, but on one "1914 Campendonk," he used a commercial white that contained titanium dioxide. Pigment analysis found the titanium white. A 1914 painting cannot contain a pigment that reached the market in the 1920s. The detail that makes the case worth telling is that the giveaway was in a small painted sketch on the back of the canvas. Beltracchi had checked the front for anachronistic materials and missed the back. That is the whole power of forensics in one anecdote. It only takes one chemical fact in one square inch.

Han van Meegeren, who forged Vermeers in the 1930s and 1940s and sold one to Hermann Göring, is the historical bookend. To fake the hard, aged surface of a 300-year-old painting, he mixed his pigments with Bakelite, a phenol-formaldehyde resin invented in the 20th century, and baked the canvases to crack them. Later chemical analysis found the synthetic resin and modern pigments inconsistent with the 17th century. The work could not be what it claimed.

What can radiocarbon dating and tree-ring analysis prove?

These two methods date the thing the paint sits on, the support, rather than the paint itself, and they are pure exclusion tools. Used correctly they establish a terminus, the earliest possible date a work could have been made, and that is often enough to end an argument.

Radiocarbon dating measures the decay of carbon-14 in organic material: the canvas fiber, the wooden panel, sometimes the binder. It returns a calendar range. The logic for authentication runs one direction. If the canvas of a supposed Vermeer dates to after Vermeer's death, the painting is not by Vermeer, period. If the canvas dates comfortably within his lifetime, that supports the attribution without proving it, because the threat runs the other way: a forger can buy a genuinely old, blank canvas at a flea market and paint a modern fake on a period support. Dendrochronology does the same job for wooden panels, matching the tree-ring pattern against regional master chronologies to find the felling date of the tree. A panel whose youngest ring dates to 1620 cannot carry a painting claimed for 1500.

The limit, again, is honest and important. These methods date the material, not the act of painting. A genuinely 16th-century oak panel could have been scraped clean and repainted last year, and the wood would still date to the 16th century. So a "pass" from carbon dating or dendrochronology removes one specific way of being fake. It does not confer authenticity. This is the recurring shape of the whole discipline. The science subtracts suspects. It does not name the painter.

Can AI and brushstroke analysis authenticate a painting?

This is the newest tool and the one most prone to being oversold, so it is worth being precise about what it does and what the numbers actually mean. AI systems, the best known being the Zurich firm Art Recognition, train neural networks on high-resolution images of securely attributed works, learn the statistical signature of an artist's brushwork and texture, and return a probability that a target work is or is not by that hand. The output is a percentage, not a verdict.

The headline results are genuinely striking and genuinely limited in the same breath. Art Recognition publicly assigned a high probability, widely reported around 90% or above, that the National Gallery's Rubens Samson and Delilah is not by Rubens, which reopened a decades-old scholarly fight. As of 2024 to 2025 the National Gallery and Rubens scholars have not accepted that as conclusive, and it stands as one piece of contested evidence in a long dispute. The accuracy figures from these systems run high on controlled tests, often quoted in the 90s, and one commercial study reported 98.2% accuracy distinguishing two closely related artists. Then reality intrudes. A 2024 to 2025 academic study that trained a model to 96.32% accuracy on telling Monet from non-Monet saw accuracy collapse to 54.04%, barely better than a coin flip, the moment it had to face AI-generated Monet-style images. The models are brittle exactly where it matters, against a sophisticated imitation designed to fool them.

Which is why the 2025 responsible-use guidelines drafted with Art Recognition and the Center for Art Law are explicit on the point: these tools produce probabilities, those probabilities do not by themselves constitute authentication, and the final attribution stays with human experts. The commercial labs say the same thing about their own chemistry. One states plainly that scientific analysis of materials "can only ever prove a painting is a forgery." That is the consensus across the entire field, old methods and new, and it is the line an investor should internalize. The instruments and the algorithms are powerful at falsification. Positive authorship remains a human judgment, assembled from science, provenance, and connoisseurship together.

What does a full forensic workup cost and when is it worth it?

A full scientific analysis at a leading commercial lab starts around $13,500 per work. ArtDiscovery, one of the major labs, quotes that figure for a package covering pigment and binder identification, molecular and elemental testing, and radiocarbon dating, with about a four-week turnaround. Museum-grade, multi-technique investigations on a complex work cluster in the $10,000 to $30,000 range and up, with imaging-only packages for less. Orion Analytical, the lab James Martin built that helped crack the Knoedler fakes, was acquired by Sotheby's in 2016.

In investing terms the math is straightforward. Set $13,500 to $30,000 against the loss if a seven-figure work turns out to be a fake, and the test is cheap insurance. The Knoedler gallery sold roughly $80 million of forged Abstract Expressionist paintings, supposed Rothkos and Pollocks and de Koonings, all painted by one man in Queens, before the science caught up. The forensic cost on the entire fraud would have been a rounding error against the losses.

We think about it the way we think about any due diligence. The science narrows risk, it does not eliminate it, and you pay for the narrowing because the downside it guards against is total. A lost attribution does not discount a work by some percentage. It can erase the value outright. Small known cost against large catastrophic loss is exactly the kind of trade that pays for itself over a portfolio, even when most individual tests come back clean.

The Bottom Line

  • Forensic science is fundamentally a tool of exclusion. It can prove a painting is a fake by finding an impossible material or date, and it cannot, on its own, prove who painted a genuine one.
  • Imaging methods read process and structure: X-radiography shows hidden compositions and construction, infrared reflectography exposes the underdrawing, UV reveals later restoration, and raking light reads brushwork and age cracks. None identifies a pigment or a date.
  • Pigment analysis is the decisive test because chemistry has a calendar. Titanium white (commercial from the early 1920s), zinc white (1830s), cobalt blue (early 19th century), and Prussian blue (1720s) each disqualify an earlier date, the logic that exposed both Beltracchi and van Meegeren.
  • Radiocarbon dating and dendrochronology date the support, not the paint, so they set the earliest possible date but cannot confer authenticity, because an old canvas can carry a new forgery.
  • AI brushstroke analysis returns a probability, not a verdict. Accuracy that runs in the 90s on clean tests can collapse toward a coin flip against a deliberate imitation, which is why the 2025 guidelines treat it as supporting evidence only.
  • A full lab workup starts around $13,500 and runs to $30,000 or more, cheap insurance against a total loss on a seven-figure work. We treat it as standard due diligence: the science narrows the risk, and the downside it guards against is the whole value of the asset.

Sources

  1. The Art Newspaper, "How the detection of titanium white gave away Beltracchi," 2012. https://www.theartnewspaper.com/2012/12/01/how-the-detection-of-titanium-white-gave-away-beltracchi
  2. Wikipedia, "Wolfgang Beltracchi" (conviction, figures). https://en.wikipedia.org/wiki/Wolfgang_Beltracchi
  3. Center for Art Law / "Tricking the Art Market on Forgery: Beltracchi and Scientific Technology." https://itsartlaw.org/cultural-heritage/wywh-tricking-the-art-market-on-forgery-beltracchi-and-scientific-technology/
  4. Hephaestus Analytical, "Imaging Techniques Explained: Pigments in Focus." https://www.hephaestusanalytical.com/blog/imaging-techniques-explained-pigments-in-focus
  5. Hephaestus Analytical, "Artificial Intelligence and the Detection of Forgery in Art Authentication." https://www.hephaestusanalytical.com/blog/artificial-intelligence-and-the-detection-of-forgery-in-art-authentication
  6. Binghamton University (Alpenglow Journal), XRF pigment study (titanium incorporated into pigments from 1921). https://orb.binghamton.edu/cgi/viewcontent.cgi?article=1166&context=alpenglowjournal
  7. Fine Art Restoration Co., "Professional Painting Assessment and Pigment Analysis." https://fineart-restoration.co.uk/news/professional-painting-assessment-pigment-analysis/
  8. CHSOS, "Infrared Reflectography and Fluorescence for Pigments." https://chsopensource.org/about-scientific-examination-of-art/chsos-application-notes/irrf-infrared-reflectography-fluorescence-pigments-checkers/
  9. ArtDiscovery (full scientific analysis pricing and turnaround). https://artdiscovery.com
  10. Center for Art Law, "AI and Art Authentication Guidelines," 2025. https://itsartlaw.org/ai-and-art-authentication-guidelines/
  11. Art Recognition, "Framework for Responsible Use of AI in Art Authentication," 2025. https://art-recognition.com/wp-content/uploads/2025/11/Framework-for-Responsible-Use-of-AI-in-Art-Authentication-final.pdf
  12. University of North Carolina at Charlotte, "Exploring AI and Art Authentication" (Monet 96.32% / 54.04% study). https://teaching.charlotte.edu/2026/02/25/exploring-ai-and-art-authentication-a-professors-perspective/
  13. Institute of Art and Law, "Handle With Care: Thoughts on Science's Role in Combating Art Crime" (Knoedler analysis). https://ial.uk.com/handle-with-carethoughts-on-sciences-role-in-combating-art-crime/

This article is for educational purposes only and does not constitute investment advice. Masterworks is a platform for investing in shares of art and does not provide authentication or appraisal services to the public. Past market behavior is not indicative of future results.