Segmenting buildings in digital surface model ( DSM) is a key technique of 3D city reconstruction based on remote sensing data.In order to overcome the poor performance caused by the negative interference of topographic relief and edge in the segmentation of buildings in DSM,a hierarchical segmentation method named LSORTSEG is proposed in light of both level-set ( LS) method and the model of occlusions of random texture ( ORTSEG) .In this method,firstly,the potential building regions are extracted roughly by means of the LS method.Secondly,the regions are expanded properly.Then,the expanded regions are segmented by using the ORTSEG for an optimized segmentation of buildings.Experimental results show that the proposed method can improve the segmentation accuracy of buildings effectively.