Abstract: Aiming at the demands of consistent descriptor and intelligent retrieval technology for 3D shapes, we propose a 3D shape analysis based on Laplacian multi-eigenmap. Firstly, we extract the surface and volumetric features of 3D models and construct a multi-feature affinity matrix based on the measurement of geometric distance, angular distance and volumetric distance. Secondly, our method converts 3D spatial domain to spectral domain by using Laplacian multi-eigenmap which effectively reveals the intrinsic invariance and consistent structure among shapes. Finally, we analyze the eigengap to adaptively determine the clustering number and implement automatic structural recognition and segmentation by combining the K-means clustering method. A series of experimental results have shown its robustness and efficiency in shape matching and shape segmentation. Our work has important significance for the high-level semantic description, shape registration and shape retrieval.