The method of extracting building boundaries based on the fusion of airborne radar (LIDAR) point cloud and aerial images is proposed by analyzing the feature of airborne LIDAR point cloud data and aerial images data. The contour line of buildings are extracted from both the point cloud and the aerial images. The contour line is fitted to lines of the building boundaries. The building vertexes are derived from two adjacent and vertical boundaries. The registration fusion of airborne point cloud and aerial images is achieved according to the correspondence vertexes of building. The point clouds get the spectral information of aerial images, which is used as a feature vector of clustering analysis to extract plants, trees and other objects. The height information is used to extract building from the buildings and roads which have similar spectral information, and the accurate boundaries of buildings are extracted and the detection of the boundaries of building is achieved. Experimental results indicate that the accuracy of point cloud classification can reach 97.96%, and the precision of the extraction of building boundaries can be up to 0.21 m, which ensures an effective way of detecting building boundaries.