Look at a photo of your friend upside down. It is suddenly hard to know who it is! Our brain reads a face as a whole, not part by part. It looks at the eyes, nose and mouth together.

When a face is upside down, this does not work well. In the Thatcher illusion, a face with upside-down eyes and mouth looks normal when the whole picture is upside down. Turn it back, and it looks strange!

Why is it so hard to recognize a face that is upside down? In 1969, a psychologist asked people to remember photos of faces and of objects such as houses and airplanes. When the pictures were upright, faces were remembered better than the objects.

But when they were turned upside down, performance dropped much more for faces. In 1980, another psychologist created the Thatcher illusion. He flipped only the eyes and mouth in a photo of a face. Upright, the face looks very strange, but when the whole picture is upside down, it looks almost normal. Researchers believe that we process upright faces as a whole, including the positions of the eyes, nose and mouth, and this system fails when faces are inverted.

Turn a photo of a friend upside down, and you may struggle to recognize the face. In 1969, the psychologist Yin compared memory for faces with memory for other things usually seen one way up, such as houses, airplanes and human figures. Upright faces were remembered better than the other objects, but turning the pictures over hurt faces far more than anything else.

In 1980, Thompson published a vivid demonstration in the journal Perception, using a photograph of the then British prime minister. When the eyes and mouth are cut out, flipped vertically and put back, the upright face looks grotesque. Turn the whole image upside down, and the oddity largely disappears. In a 2008 review, the Belgian researcher Rossion argued that upright faces are processed holistically, with features and their spatial relationships combined into a single representation.

Inversion breaks this process and forces a slower, feature-by-feature analysis. In his view, the change is one of kind, not merely of degree. Some researchers disagree, maintaining that weaker holistic processing survives in inverted faces, and Yin's original experiment was small. Paradoxically, the weakness grows out of expertise. We see countless upright faces, so the brain becomes highly specialized for them.