DescriptionJudgments of similarity play an integral role in the human cognitive system as they provide a means for extracting information about how objects in the world relate to each other. This similarity information is applied in various cognitive tasks, such as categorization, recognition, and identification. Previous work suggests that perceived objects are cognitively represented in a psychological space where similarity is preserved, allowing for an internal structured representation of objects in the world (Shepard, 1964). For an internal representation to be formed, information about an object must be extracted. Shape, a highly informative and salient property of an object, is often used. Judgments made about shape similarity reflect how humans functionally represent and utilize shape information from an object. Computational shape representation has been achieved with varying amounts of success (e.g. Blum, 1973; Biederman, 1987). This variability is due, in part, to the complexity of mimicking the seemingly effortless human ability to make judgments about shape even in spite of numerous possible complications, such as sparse information and occlusions. This work presents the use of a Bayesian estimation of a shape's skeleton, the maximum a posteriori (MAP) skeleton (Feldman & Singh, 2006), as part of a generative model of shape that allows for the computation of a probabilistically-based similarity metric. This method of shape representation makes possible the prediction of similarity judgments reported by human subjects on collections of shapes that exhibit differences in both part structure and metric qualities and that have been generated by an unrelated process. It is argued that the derivation of a similarity metric from this model provides the previously unavailable relationship between shape representation and categorical judgments about shape.