The nature of the processes underlying face perception was examined in two different paradigms using the same set of stimuli varying on three dimensions of two values each. In a simultaneous matching task, both latencies and errors were found to decrease as the number of differences between stimuli increased. Regression analyses showed that the manipulated features interactively contributed to these variations when the faces were presented in their normal upright orientation, whereas no evidence of an interactive processing was found when the faces were inverted. A multidimensional scaling analysis of dissimilarity judgements between pairs of different upright faces revealed that the overall impression of a face was not simply the sum of subimpressions, and that a configuration typical of each face emerged from the relationship among their particular features, giving each face its individuality. The results suggest that faces have both component and configural properties and lend themselves to different processing strategies that are not mutually exclusive and can unfold simultaneously. Some implications of these results for current research on face perception and recognition are outlined.