With the developments of 3D (Three-dimensional) modeling and virtual reality technique, 3D surface classification becomes a fundamental problem in computer graphics and computational geometry. We focus on classifying a 3D surface into rough vertexes which compose intensity change degree parts and the smooth vertexes which are in plain sections. Inspired by Canny edge detection algorithm and multi-scale method, we proposed a multi-scale classification algorithm for 3D surface classification which is useful for compression and retrieval of 3D model. The results shows 3D surface can be classified accurately in different scales with the algorithm.