A K-means clustering method is proposed based on Gaussian map. First, we search the neighboring points of the target points using a k-nearest neighbor search. Gaussian map is then performed on the normal vectors of the set of triangles consisting of the target point and its neighbors. The silhouette coefficient is selected as the cluster validity index to determine the optimal cluster number. According to the clustering distribution of different surfaces, the feature lines of the three-dimensional laser point cloud model are obtained. The experimental results show that the proposed evaluation index is easy to use and has less noise than other indexes. It can extract the feature lines of regular and irregular point clouds completely and efficiently.