For 3D laser point cloud scanned by lidar on power lines, an automatic method was designed to divide point cloud into 3 types: point of lines, point of pylons, and point of ground objects. This process can lay essential foundation for spatial distance check in power transmission system. The core of the method is making full use of global statistical features of point cloud instead of local features, thereby maximally prevented limitation of the algorithm. According to dozens of practical tests from State Grids, the automation and universality of this method are satisfactory, which largely reduce the work usually laborious in traditional classification with commercial softwares by half-automatical interactive process, and shorten the average processing time from about 1 hour to several seconds, greatly improve working efficiency and userexperience.