How Experts Check If an Old Photo Is Fake

Experts check an old photo by first tracing where it came from. A reverse image search is the usual first step, because it can turn up an earlier or original version. They then test the picture itself: whether light, shadows and reflections agree, whether any region has been copied and, for digital files, whether the camera's own traces are intact. No single test settles the matter. The 1917 Cottingley fairy photographs show why.

This article sets out what the main checks look for, then follows the Cottingley case, where the expert opinions of the day pointed in different directions and the truth came from the people who made the pictures. For a wider method, see the practical checklist for evaluating a claim.

Start by tracing where the picture came from

The first step is usually to find out whether the image exists anywhere else. A reverse image search can locate an earlier or original version of a picture that is being shared.

Hany Farid, a professor of computer science at Dartmouth who works on photo forensics, notes that people are not very good at spotting manipulated images by eye alone, so a visual impression is not enough on its own.

An earlier copy with a different caption or date is a strong lead, but it does not prove the picture is genuine. It only shows that the story attached to it needs testing.

Read the light, shadows and reflections

Lighting is one of the main tests for composite images, where several photographs have been joined. Software can estimate the direction of the light falling on each person or object, and if the directions disagree the image was probably assembled from separate shots. In one police-and-ducks example the light directions did not match, but other items would need checking before concluding that the ducks had been added.

Shadows should fall consistently with the light source. Farid notes that this check works only when a point on a shadow can be clearly matched to a point on the object casting it.

Reflections in the eyes give the position of the lights in the scene. In one publicity photograph of a television show cast, the highlights were so inconsistent that a visual check was enough to show it had been doctored.

Copy-and-paste cloning is another common form of manipulation. In a test on a 2004 political advert, an algorithm found three identical regions.

Check the digital file and its traces

For a digital original, the camera's colour filter leaves a pattern of pixel correlations that should match that camera. If the pattern is missing in a small area, that area may have been changed. The method only works on an original digital file, because a scan of a printout adds new correlations of its own.

Metadata, or EXIF data, can hold camera settings and possibly GPS location. A software tag naming an editing program does not prove deception, since many ordinary photos receive routine edits. Image dimensions can be compared with the camera's typical output to suggest cropping, and motion blur should match the shutter speed used.

Farid suggests brightening dark areas with Photoshop's Levels tool, which can reveal material that was erased. He also warns against zooming in beyond 200 or 300 per cent, because at higher magnification lens artefacts can look like manipulation.

The Cottingley case as a test of expert opinion

The Cottingley Fairies photographs were five images taken in 1917 by Elsie Wright, who was 16, and her cousin Frances Griffiths, who was 9. Elsie's father, Arthur, a keen amateur photographer with his own darkroom, dismissed the first image as cardboard cutouts, partly because of his daughter's artistic ability. The full story is told in our piece on the Cottingley hoax and Conan Doyle.

Edward Gardner, who championed the pictures, had secretly marked the plates to check for tampering. He sent the prints and glass-plate negatives to Harold Snelling, who said the negatives were entirely genuine and showed no trace of studio work with card or paper models.

Gardner and Doyle asked Kodak for a second opinion. Kodak technicians said the pictures showed no signs of being faked but that this could not be taken as conclusive evidence, and the company declined to issue a certificate. Ilford reported some evidence of faking. Gardner and Doyle counted the three opinions as two in favour and one against, which shows how a verdict can depend on how the opinions are weighed.

Oliver Lodge believed the photographs were fake and doubted the hairstyles. Later investigations revealed that Gardner had asked Snelling to improve the negatives, which weakens the value of the earlier opinions.

What finally settled it, and what remains unclear

In the early 1980s, some sixty years later, the girls admitted the photographs were faked. Elsie had drawn and cut out the fairies and attached them with hatpins, using cardboard figures copied from a popular children's book.

Elsie said a double exposure had been used for the fairy bower picture, but Frances insisted it showed real fairies. Frances also maintained that the fifth photograph was genuine. The research behind this article did not establish whether that point has since been settled.

The photographs and cameras were held at the National Science and Media Museum in Bradford (as of 2019). In 2024, heritage scientists at the University of Bradford used CT scanners to examine the cameras, a reminder that physical objects can be tested in ways that prints cannot. The findings of that work are not drawn on here.

The main lesson is that the strongest evidence came from the people who made the images and later explained how they did it, combined with examination of the original objects, rather than from the expert verdicts of the time. Other famous cases, such as the Surgeon's Photograph of Loch Ness and the crop circles hoax, also show how a picture can be tested against its origin. You can test your own eye in Real or Rigged?.

Why a single check is never the final word

Each method gives evidence, not a verdict on its own. A reflection that does not match, or a missing camera pattern, is a reason to look harder, and a clean result on one test does not rule out editing elsewhere in the picture.

Scans of old prints lose the pixel traces that the camera-fingerprint method depends on, so the most useful digital tests usually cannot be applied to a family print or a scanned copy. The checks work best together and against a stated claim: where the photo came from, who took it, what was said at the time, and whether the explanation fits the evidence.