Abstract: Combining multiple shape descriptors can greatly improve retrieval accuracy.However,it will decrease retrieval efficiency due to the growing feature dimension.This paper addresses this problem,and proposes a new method to combine multiple descriptors,namely mean descriptor.It can greatly improve retrieval accuracy while remaining retrieval efficiency.Firstly,the proposed algorithm uses spherical harmonics to generate harmonics descriptor.Secondly,it constructs nearest graph for models by employing multiple descriptors.Finally,with the above harmonics descriptor and nearest graph,the mean descriptor is generated by meaning operation.The experiments on a public model database benchmark show the retrieval accuracy of mean descriptor has a great improvement over that of harmonics descriptor.Even comparing with the existing multiple shape descriptors,the retrieval accuracy has a small improvement.