Visual object recognition could rely upon dissociable or shared mechanisms. Domain-specific accounts argue that object categories like words and faces are processed by largely independent mechanisms. On the other hand, some accounts assume that face and word recognition share or even compete for the same cortical resources. We assessed performance with faces, houses, and pseudowords in a task where either features or configuration of features were systematically varied/manipulated (N = 101). On each trial of the task, a sample (unfamiliar face, unfamiliar house, or pseudoword) appeared at screen center followed by match and foil images displayed simultaneously to the left and right of screen center. The match image was identical to the sample image, but the foil image was different either featurally or configurally. To estimate the separability of visual object recognition mechanisms, we used representational similarity analysis (RSA) where a correlational matrix for accuracy (reference model) was compared to predicted data patterns. By means of Bayesian regression, we found that a face specialization model – where processing of faces differs from processing of both words and houses and where processing of houses and words does not differ – provides the best solo account of the results. This suggests that faces are processed in a specialized manner in this task while objects and words rely on common processes.