A 3D digital image model is proposed to represent the LIDAR data. The mathematical morphology is extended to 3D and then, dilation and erosion operators are re-defi ned. A method combining 3D mathematical morphology with clustering analysis is developed . Sequential dilation operations and clustering analysis are introduced into the 3D point cloud to achieve the pixel- level results of point cloud. The relationships between the two parameters and data property, resolution of point cloud and the minimum distance between objects, is discussed. Two case data are used to demonstrate the feasibility of the proposed method. The result for the fi rst dataset is compared with those from the two other methods, Mean Shift algorithm and adaptive TIN fi lter method. The advantages and disadvantages are summarized using segmentation evaluation factors, segmentation accuracy, and computation effi ciency. Meanwhile the stabilization of proposed method is also analyzed.