Abstract: Generalization of road network is one of the focuses in map generalization, while selective omission is the most important problem in road network generalization. In recent years, many solutions have been proposed for road selective omission, but how to maintain the overall and key local structures of original networks has not been solved yet. Given this, an auto-selection method considering structural, and geometric properties of road networks is proposed. Degree of centrality, clustering coefficient and geometric length were taken into account in this approach. The method is based on dual graph generated by the strokes of road networks and then stroke length, degree of centrality and clustering coefficient were derived by means of network analysis. Feasibility and reliability of this approach were examined by a road selection process at different scales through comparing with benchmarks. An experiment shows that at given scales, roads selected by this approach were consistent with those selected by cartographers. This proposed process performs well not only in maintaining the global and local structures of the road networks, but also in keeping the topological structure in terms of road network connection from a large scale map to a small one. In summary, the method was proved to be stable and reliable.