An automatic classification method for urban area lidar point clouds is proposed. A segmentation algorithm named topology heuristic segmentation to process regular grid height image is used. A separation for terrain objects and off-terrain objects is carried out with Otsu clustering algorithm and adjacent regions is merged. The typical off-terrain objects could be detected by using topological models, the tree can also be detected by multi-echo ratio of region. An experiment is done with a typical urban cloud points, the result shows that the algorithm has good classification accuracy and strong practical value.