This study investigates the rules by which the component features of faces are combined when presented in the left or the right visual field, and it examines the validity of the analytic-holistic processing dichotomy, using concepts elaborated by Garner (1978, 1981) to specify stimulus properties and models of similarity relations as performance criteria. Latency measures of dissimilarity, obtained for the two visual fields, among a set of eight faces varying on three dimensions of two levels each, were fitted to the dominance metric model, the feature-matching model, the city-block distance metric model, and the Euclidean distance metric model. In addition to a right-visual-field superiority in different responses, a maximum likelihood estimation procedure showed that, for each subject and each visual field, the Euclidean model provided the best fit of the data, suggesting that the faces were compared in terms of their overall similarity. Moreover, the spatial representations of the results revealed interactions among the component facial features in the processing of faces. Taken together, these two findings indicate that faces initially projected to the right or to the left hemisphere were not processed analytically but in terms of their gestalt.