To registration problem of scanned point clouds data without any additional information, a novel normal vector based automatic registration algorithm is proposed. The feature points are extracted according to the change of local normal vector, and the initial matching points are found through histogram feature proposed in this paper. The random sample consensus (RANSAC) is used to get the accurate matching points according to the distance restriction. The initial registration parameters are computed by the quaternion, and iterative closet point (ICP) algorithm is used to get accurate result. The experimental results show that this algorithm is effective.