The extraction of the building models from remote-sensed data is an important work in the geographic information systems. This investigation describes an approach to integrate LIDAR data and 2D building boundaries for building reconstruction. The proposed scheme comprises of three major parts: (1) data pre-processing, (2) extraction of building primitive, and (3) shaping. In the data pre-processing, the LIDAR point clouds are structuralized by Triangulated Irregular Networks (TIN) and interpolated as Digital Surface Model (DSM). The building boundaries are traced to form closed polygons. In the extraction of building primitive, we use DSM with Canny Edge Detector to determine the step edges. The ridge lines are extracted from the raw point clouds by using a TIN-based region growing technique. The coplanarity and adjacency are considered for the TIN-based region growing. Finally, the building polygon is divided into several building primitives by the extracted structure lines. In the building reconstruction, we shape the roof of each building primitive by using LIDAR point clouds. The proposed method has been tested with LIDAR data acquired by Leica ALS50 and 1/1000 topomap. Experimental results indicate that the proposed scheme reaches high reliability, and the shaping error of the reconstructed models is better than 0.20 meter.