Abstract

Face recognition is often believed to be based on some sort of ‘‘configural’’ or ‘‘holistic’’ processing. Although these terms are often not well defined in the literature, an influential hypothesis is that the recognition of individual faces depends, to a considerable extent, on shape in terms of metric distances between features – so-called “second-order configural information”. Here I will argue that this popular idea is in striking contrast to a number of recent empirical findings, and therefore is not only simplistic, but also wrong. First, I will show evidence that familiar faces are well recognized even when idiosyncratic shape has been eliminated by shape normalization – suggesting a strong role of surface reflectance/texture information for recognition. I will then present several face learning experiments that examined the impact of selective photorealistic caricaturing of either shape or reflectance information on face recognition, and on its neuronal correlates in event-related brain potentials. These experiments confirm that shape is of little importance for the recognition of familiar faces, whereas shape enhancement can facilitate the encoding of new faces. Moreover, distinctive reflectance information was found to be particularly important for the recognition of learned faces. Overall, the present experiments underline the importance of face familiarity for mental representations of faces. The configural processing hypothesis in its traditional form does not well account for the findings reported, and thus needs to be reconsidered.