In this paper, we propose a 3D shape similarity measurement method using the surface skeleton of a voxelized 3D shape. A set of features are extracted from the skeleton by combining the skeleton and the distance map of the subject. The distribution of these extracted features is used to represent the 3D shape. Two 3D shapes can be compared with a similarity score computed from the distance between their respective features distribution. Our method is robust to noise and to partial occlusion. Moreover, it is scale and rotation invariant. We tested and validated our method with various objects with different shape and topology.