To realize 3D CAD model reuse requirement of multi-granularity, precise and intelligent in design field more efficiently, a method of typical structure mining in 3D CAD model and similarity evaluation based on simulated annealing algorithm was proposed. Firstly, attribute adjacent graph of CAD model was established through extracting topological and geometrical information from the 3D CAD model represented by B-rep, which can record face and edge detailed structures of 3D CAD model and its organization and association. Then, according to its description carrier, association graph and association graph matrix were established with the mapping relation of vertex and edge in attribute adjacency graph between typical structure and CAD model, and a mechanism of similarity estimation was proposed at the same time. At last, typical structure mining in 3D CAD model was accomplished based on simulated annealing algorithm through excavating maximal clique in association graph, while similarity estimation between typical structure and return model was accomplished. Using general model base, agricultural machine and equipment model base, algorithms testing and validating were completed compared with ant colony algorithm and genetic algorithm. Experimental results showed that the proposed method can realize typical structure mining in 3D CAD model and similarity estimation, which can effectively support design reuse of feature or local structure in 3D CAD models design domain.