A novel approach for examining shapes as shells in a parametrical space is described. The parameterical space is defined in terms of shell radius and sets of angular extensions. The representation reflects the global and local characteristics of shape. These template shells are compared to the given clusters of shells which are extracted from known models. The comparison is performed simply by calculating the similarities between the corresponding extensions. It is demonstrated that the shell descriptors are independent of size, rotation, and displacement. The proposed technique is being applied to different shapes and has the ability to process partial recognition for occluded and overlapped shapes.