Résumé : The focus of medical image registration has shifted to nonrigid registration. It is also a core component in computational neuroanatomy. In this paper, we propose an efficient registration framework which has feature of multiresolution and multigrid. In contrast to the existing registration algorithms. Free-Form Deformations based NURBS (Nonuniform Rational B Spline) are used to acquire nonrigid transformation. This can provide a competitive alternative to Free-Form Deformations based B spline on flexibility and accuracy. To our knowledge, it is the first report of the use of this mathematical tool for medical image registration. Subdivision of NURBS is extended to 3D and is used in hierarchical optimization to speed up the registration and avoid local minima. The performance of this method is numerically evaluated on simulated images and real images. Compared with the registration method using uniform Free-Form Deformations based B spline, our method can successfully register images with improved performance.