A multi-level urban spatial pattern recognition method based on road network skeleton line was proposed to reveal urban hierarchy and complexity in order to address the limitation where existing research on road network and building group was often confined to single-level analysis. A road network evaluation system was constructed by integrating space syntax and fractal geometry in order to extract macroscopic skeleton lines and analyze their topology. Road mesh was automatically extracted via morphological quantification and adjacency analysis. Spatial pattern of road mesh was identified by using the minimum spanning tree (MST) algorithm. Building was partitioned based on mesoscopic skeleton line. Automatic extraction and pattern recognition of building group were achieved by using hierarchical density-based spatial clustering of application with noise (HDBSCAN), Prim’s MST algorithm and the C4.5 decision tree. Validations in Shanghai, Chengdu, Wuhan and Lanzhou demonstrated the effectiveness of the method. The method can be used to efficiently and accurately identify urban spatial pattern across different levels.