This paper presents a novel global object descriptor, achieving a balance of descriptiveness, robustness and efficiency. The proposed descriptor forms a comprehensive description of an object instance by encoding projection statistics in terms of contour signature and distribution matrix (CSDM). To generate a CSDM descriptor, a local reference frame is defined to align the object's point cloud with the canonical coordinate system. After that, the sub-histogram of contour signature and distribution matrix can be determined from orthographic 2D projected patterns. Finally, a CSDM descriptor is generated with a concatenation of sub-histogram. In recognition stage, a two-stage comparison metric is designed to eliminate information redundancy. A comprehensive performance evaluation in terms of scalability, descriptiveness, robustness and efficiency is performed on the publicly available dataset. Experimental results show that the performance of CSDM descriptor is comparable with the other two state-of-the-art descriptors.