What is it about the way faces are represented by the visual system that makes them so much harder to recognize when inverted? The authors tested the hypothesis that the "face inversion" effect results from the use of holistic shape representations. This suggests that the susceptibility of nonface patterns to inversion should be a function of their degree of part decomposition. In Experiment 1 this was tested and confirmed with dot patterns in which the degree of part decomposition was manipulated by grouping and segregation on the basis of dot color. The hypothesis also predicted that the face inversion effect can be eliminated with face stimuli if participants are induced to recognize the faces in terms of their component parts. In Experiment 2 this was tested and confirmed with whole, intact faces, in which the degree of part decomposition was manipulated by allowing participants to study them, initially, in either whole, intact versions or versions with parts presented separately.