Abstract: Based on the existing post-processing methods of LiDAR data, this paper proposes a new“separated step-by-step”filtering method of point cloud. First, a“rough”filtering method is applied to the LiDAR point cloud and the“ground points hypothesis”and“non-ground points hypothesis”are gained. Then, a causal auto-regressive model (car-model) is imported to do modeling of the ground surface and hypothesis test for the two classes of point clouds, and ground points and non-ground points are classified by the results of the hypothesis testing. Finally, a reliable classification results is gained. Compared to the“Least-Squares Prediction Method”and“mathematical morphology”, the results of LiDAR point cloud filtering by the“separated step-by-step”processing method is more reliable.