Abstract: Topology structures are critical for 3D CAD models, which are described in nonlinear features such as graphs or trees. However, the existing clustering algorithms cannot cluster these kinds of nonlinear features effectively. Aimed at this situation, this paper proposes a nonlinear feature oriented 3D CAD model clustering algorithm. Firstly, various nonlinear features are characterized as attribute graphs uniformly, and the distance matrix of attribute graphs sequence is defined; secondly, with the distance matrix as input, a nonlinear agglomerative hierarchical clustering algorithm is put forward to cluster the attribute graphs; finally, using the clustering results as learning samples, a dynamic classification algorithm is introduced to classify the new added graphs. The reusable regions of 3D CAD models are clustered effectively based on the above algorithm. Theoretical analysis and experimental results demonstrate the effectiveness of the proposed algorithm.