The efficiency of the extracted buildings always is a key in engineering application of LiDAR points. Aimed at the problem that the efficiency of filtering first and then extracting methods, currently in effect is low, a method combining Delaunay TIN models and region growing for extracting buildings from raw LiDAR data is put forward in this paper. Firstly, Delaunay TIN models are built on the original LiDAR points. Edge points of buildings can be extracted by using the normal vector, length of side and point height of triangles where the edge points are located. Then, the extracted edge points are assigned as seed points in order to implement region growing based on triangle network connections which will yield a points set of protrusion. Finally, since the number of non-building points is much smaller than that of the building points, the non-building points set can be deleted while the building points set is reserved. The method in this paper can be used to extract building points set and edge points directly without the operation of filtering and provide foundation for further contour extracting and building reconstruction. The simulation results show that the method has obvious efficiency under the guarantee of accuracy in extraction and has a certain of adaptability.