This paper presents a fully automatic method for generation of 3D building skeleton models. Objects are reconstructed from point clouds, without the need for a support, like additional information and data sets. This condition makes the reconstruction task even more complicated, however at the same time, presented algorithm becomes much more versatile. Topological model of a building is created based on characteristic points, which determine intersections of adjacent planes of the roof or walls. The characteristic points located on the outer edges of a building are extracted using author’s contour detecting algorithm. In the next stage of reconstruction topological relations between the points are defined, which allow to detect contour lines of individual planes of a building. Finally, adjustment of vertex points and edge lines is performed that enables to obtain regularized building model. The algorithm was tested against airborne laser scanning data set that shows a part of the small town. As a result of experiments it can be concluded that the described method allows the correct generation of skeletal building models with a complex structure.