Radius Angle Histogram (RAH) is useful method of retrieving 3D mechanical models. This method, however, not completely uses the information of radius angle histogram of the model. As a result, the retrieval precision is not very high enough. To improve the retrieval efficiency, K-means clustering method (KMM) is proposed in this paper. The radius angle histograms of the models are established first and then be served as inputs as KMM, respectively. By using KMM, the models can be classified and the results can be obtained. To validate the proposed method, an experiment is given. The results show that the retrieval precision of the proposed method is higher than that of single using RAH method.