Abstract
Exposure to a face can alter the appearance of faces viewed subsequently. For instance, a male face can make subsequent androgynous faces look female. It has been proposed that these aftereffects are driven by a ‘face space’ opponent neural code (Leopold et al., Nature Neuroscience, 2001). The centre of this ‘face space’ is thought to reflect a time average of encountered faces. Exposure to a distinctive face is thought to shift the average toward that face, making previously average looking faces take on the appearance of a face that differs from average in the opposite manner to the initial face - hence an androgynous face looking more female after exposure to a male face. Here we present data that, we believe, pose a fundamental challenge to this account. We simultaneously adapt people to male and female faces in different positions, causing androgynous faces to look more female and male respectively. These effects persist when initial and subsequent faces differ in size, which is usually taken as evidence ruling out the contribution of low-level retinotopically organized coding mechanisms. The existence of simultaneous, opposite facial aftereffects is inconsistent with mediation via a single opponent facial code. Our data are consistent with there being multiple, spatially localised, opponent facial codes. However, we cannot see what functional benefit this arrangement would subserve. Instead, we suggest our data, and facial aftereffects in general, are explicable via a combination of low-level spatial aftereffects (that exaggerate differences in position and orientation) and a bias to categorize indistinct inputs as being opposite to immediately preceding distinctive inputs.