Abstract: In this paper,a CAD model retrieval method based on inexact graph matching is presented in order to resolve the problem that the exact graph matching is unable to support the similar model retrieval.First,a representation of face attributed relational graph(ARG)for each CAD model is extracted from its B-rep model.Then,the vertex compatibility matrix and edge compatibility matrix between the ARGs of the target and searched model are calculated,and the measure of the similarity between the two models is created from the compatibility matrices,which serves as the objective function for optimally selecting vertex mapping matrix M between the two models.Finally,the optimal vertex mapping matrix M is found using Sinkhorn's alternative normalization method for M's rows and columns after relaxing M's elements to be continuous.Experimental results show that this method is able to support the inexact model retrieval and its efficiency meets the requirement of practical applications.