This thesis presents work on several aspects of 3D shape processing. We develop a learning based surface reconstruction algorithm that is robust to typical input artefacts and alleviates the restrictions imposed by previous such methods. Using the human shape perception motivated paradigm of representing a 3D shape by its 2D views obtained from its view sphere, we compute the shape's "best views", extend these views to obtain the more dynamic "best fly" of the shape and also compute shape complexity which is used to compare the shape with others so as to obtain an ordering. Our example based method to "correctly" reorients a 2D shape in an image is also presented as well as a strategy to approximate shape descriptor values on the view sphere using just a few samples This allows to bypass the often time consuming requirements of evaluating the descriptor on a dense sampling of the view sphere to obtain an accurate representation. We also present our work on accelerating shape similarity retrieval by using techniques from text retrieval. Lastly, we present some of the guiding principles behind the maintenance and development of a large scale, publicly accessible shape respository.