The 3D position of ground point can be precisely determined using airborne scanning laser ranging system. An algorithm for extracting man-made buildings from sparse sampled laser points acquired by airborne 3D imager is described. According to the height similarity and neighbor property of the laser sampled points belonging to the same building, the laser sampled points of the same building are clustered and segmented firstly. The boundary laser points near to the contour of the buildings are then separated based on the azimuth of laser points corresponding to the predicted heading. The separated boundary points are adjusted according to the character of man-made building, which usually forms a regular polygon. In regulating the edges of a building, the boundary points are grouped firstly based on the azimuth change. The main direction of the building is then calculated using the grouped points and least-square principle. According to the main direction and grouped boundary points, the regular edges of the building can be determined precisely. So the contour of the building can be extracted. Some buildings are successfully extracted using the algorithm and a test result is given.