Abstract: To realize feature-based reverse engineering,machining features are extracted from the triangular meshes by transforming the Morse-Smale complex into the attributed adjacency graph.First,a curvature-character indicator function is defined and the curvature of each vertex is calculated based on the moving least-square surface.The calculation of curvature is not influenced by connection relationship of the triangular meshes and can reduce noise of input point data.Then,a Morse-Smale complex is constructed and simplified on the triangular meshes to extract feature lines accurately and to segment the mesh into connected regions with definite boundaries.Finally,according to the attributes of critical points and feature lines,the Morse-Smale complex is directly transformed into the attributed adjacency graph to recognize machining features.The algorithm is executed automatically,effectively and efficiently and achieves satisfying experimental results on complex models with noise.