The relative dominance of componential and configural information to face processing is a controversial issue. We investigated this issue by examining how componential information and configural information interact during face processing, using Garner's speeded classification paradigm (Garner, 1974). This paradigm examines the ability to process one dimension of a multidimensional visual stimulus, while ignoring another dimension, using selective attention measures, and provides a powerful test of perceptual separability between stimulus dimensions. When classifying upright faces varying in components (eyes, nose, and mouth) and configural information (inter-eyes and nose-mouth spacing), observers were unable to selectively attend to components while ignoring irrelevant configural variation, and vice versa (indexed by symmetric Garner Interference). Performance with inverted faces showed selective attention to components but not to configural information (indexed by asymmetric Garner interference). When faces varied only in components, spatially distant or spatially close, selective attention to different components was possible (nearly zero Garner interference). These results suggest that facial components are processed independently, and that components dominate the processing of inverted faces. However, when upright faces vary in componential and configural information, as in natural faces, the processing of componential information and the processing of configural information are interdependent, with no necessary dominance of one type of information over the other.