Abstract: Due to complexity of conventional graph matching and sensitivity of graph seriation to its topology, this paper proposed an approach to B-rep model retrieval based on similarity of ordered paths.At first, facets of the model are sorted by their properties.Then the facets are taken as the nodes of paths, and the facets with same or different property are chosen to form ordered paths of homogeneous or heterogeneous facet respectively, which is taken as a linear representation of B-rep model.Based on this representation, a distance measurement between B-rep models which is calculated from the distance of ordered paths is presented.By calculating distance of two ordered paths and distance of ordered paths sets in a stepwise manner, the distance between B-rep models is determined.The results from searching experimental tests demonstrate that the approach based on ordered paths has robustness against variation in topological structure and the retrieval accuracy is better than that of existing methods.In addition, the time for online matching of this approach is with polynomial-time complexity.