The data of Light Detection And Ranging（LiDAR）is a discrete threedimensional point cloud, and the 3D laser foot points belonging to the same building have the characteristics of random distribution. Owing to the varied shapes of building, and the complex environments around building, the extraction of building borderlines becomes more difficult. In extracting borderline, the most important is to extract the edge point. In the paper, a new algorithm of extracting the edge point is presented. Firstly, a specific radius and threshold are set. Secondly, taking the first point as the centre of a circle, a pack circle is built. Thirdly, the distances between the centre and other points are acquired, and the number in the pack circle is counted. And then, the operation throughout all point in the point cloud is done repeatedly. Finally, the point number in every pack circle is compared with threshold to extract the edge point. Compared to the alpha shape algorithm, this algorithm can reduce the running time on the basis of keeping the extraction effect of edge point, and can improve the overall effect obviously.