This paper describes a process for the automated generation of 3D buildings using 2D building footprints derived from cadastral maps and LIDAR point cloud data. In our approach we extract relevant geometric information from 2D building footprints in order to classify point cloud data. One key concept is the fact that roofs in most cases are aligned to the angles of the walls of a building. This concept is utilized to create contiguous surfaces and to extract ridges. In a field study involving two historic city centers in Switzerland we evaluate the results of our approach.