The three-dimensional shape descriptors have been applied to object recognition and classification. A novel three-dimensional shape descriptor combined with local and global representations is proposed in this paper. Firstly，traditional moment invariants are extended by including a term to represent spatial curvature and a series of new moment invariants named spatial curvature moment invariants is constructed. Secondly，normalization method of these invariants is presented and they are independent of the translation，rotation and scaling transforms. Experiments indicate the proposed method is of lower computation complexity than traditional three-dimensional shape descriptors without reducing the recognition rate.