In this paper we present a three-dimensional model retrieval system. In our approach, a three-dimensional model is described by three invariant descriptors: a curvature index which consists of a histogram of the principal curvatures of each face of the mesh, a histogram of distances between the faces, and a histogram of the volumes based on each face. This work focuses on extracting these invariant descriptors from the three-dimensional models, and on combining these descriptors in order to improve retrieval performance. An experimental evaluation demonstrates the satisfactory performance of our approach on a fifty three-dimensional models database.