It is widely known that the road network layout can impact the non-motorized users’ traffic safety by changing the non-motorized traffic volume and road users’ behavior. Different road network patterns lead to different traffic safety levels for non-auto users and a single pattern can even have both the safe and unsafe features at the same time. By knowing what features can lead to safer traffic environment, existing road networks can be improved and new network patterns can be produced by combining all safe features from different patterns. Therefore, the associations between road network structure and pedestrian-bicyclist crashes are analyzed in this paper to determine how the structural features of a road network affect non-motorist safety. Three structural measures including average geodesic distance, network betweenness centrality, and overall clustering coefficient are calculated based on the road networks of 321 census tracts in Alameda County, California. Then the three measures together with other factors like traffic behavior, land use, transportation facility, and demographic features are employed separately in a spatial statistical model called geographically weighted regression. Conclusions are: if a network is more highly centered on major roads, there will be fewer non-motorist crashes; the network which has more average number of intersections between each pair of roads tends to have fewer accidents for pedestrians and bicyclists; and, the more a network is clustered into several sub-core networks, the lower the non-motorist crash count will be.