In the process of part design, efficient and accurate part similarity retrieval is an important way to realize effective design reuse and knowledge reuse. This paper proposes a part similarity calculation strategy based on Ullmann algorithm optimization. Firstly, a complete part complex network model is constructed to represent model information with a smaller scale descriptor and provide accurate mathematical model for subsequent similarity calculation. Secondly, the Ullmann is pruned according to the part information to solve the isomorphic graph between large-scale networks, and the similarity solution corresponding to the geometric information and the component similarity calculation to the public subnetwork to obtain the final multi-feature fusion similarity calculation based on the subgraph. The example shows that this method achieves good results in the process of complex parts information description and similarity calculation, improves the retrieval efficiency and provides reliable knowledge reuse basis for the development of new parts.