Recent neuroimaging studies have described a differential activation pattern associated with specific object images (e.g., face-related and building-related activation) in human occipito-temporal cortex. However, it is as yet unclear to what extent this selectivity is due to differences in the statistics of local object features present in the different object categories, and to what extent it reflects holistic grouping processes operating across the entire object image. To resolve this question it is essential to use images in which identical sets of local features elicit the perception of different object categories. The classic Rubin vase-face illusion provides an excellent experimental set to test this question. In the illusion, the same local contours lead to the perception of different objects (vase or face). Here we employed a modified Rubin vase-face illusion to explore to what extent the activation in face-related regions is attributable to the presence of local face features, or is due to a more holistic grouping process that involves the entire face figure. Biasing cues (gratings and color) were used to control the perceptual state of the observer. We found enhanced activation in face-related regions during the “face profile” perceptual state compared to the “vase” perceptual state. Control images ruled out the involvement of the biasing cues in the effect. Thus, object-selective activation in human face-related regions entails global grouping processes that go beyond the local processing of stimulus features.