What's lost in inverted faces?

Disproportionate inversion decrements for recognizing faces and other homogeneous stimuli are often interpreted as evidence that experts use relational features to recognize stimuli that share a configuration. However, it has never directly been shown that inversion disrupts the coding of relational features more than isolated features. Here we report three studies that compare inversion decrements for detecting changes that span the isolated-relational features continuum. Relatively large inversion decrements occurred for relational features (Thatcher illusion changes, internal feature spacing), with smaller decrements for isolated features (presence/absence of facial hair or glasses). The one discrepancy was a relatively large inversion decrement for detecting changes to the eyes and mouth, which we had classified as an isolated feature change. However, this decrement disappeared when the features were presented out of the face context (Experiments 2 and 3), suggesting that it occurs because subjects spontaneously code relations between the features and the rest of the face. Although the results support the interpretation of disproportionate inversion effects as evidence of relational coding, the difficulty of classifying changes as isolated or relational highlights an undesirable ambiguity in the isolated-relational feature distinction. We therefore consider alternative construals of the configural coding notion.