A new approach for extract and segment building target from terrestrial laser scanner point clouds is presented. The local geometric features (optimal radius, normal, dimensional feature) of each point are calculated by principal component analysis of progressive radius. The points of ground are removed from original point clouds by geometrical characteristic, so that the rest points can be divided into several point-clusters by distance. The statistics features of each point-cluster are calculated to extract the buildings. A growing rule based on the local geometric features is made to segment the buildings into surfaces. The experimental results show that the proposed method has the ability to extract and segment building form wide scene. Besides, the stability and accuracy of building segmenting by the proposed method is higher than that by traditional region growing method.