The registration of 3-D objects is an important problem in computer vision and especially in medical imaging. It arises when data acquired by different sensors and/or at different times have to be fused. Under the basic assumption that the objects to be registered are rigid, the problem is to recover the six parameters of a rigid transformation. If landmarks or common characteristics % between both objects to register are not available, the problem has to be solved by an iterative method. However such methods are inevitably attracted to local minima. This paper presents a novel iterative method designed for the rigid registration of 3-D objects. Its originality lies in its physical basis: instead of minimizing an energy function with respect to the parameters of the rigid transformation (the classical approach) the minimization is achieved by studying the motion of a rigid object in a potential field. In particular, we consider the kinetic energy of the solid during the registration process, which allows it to «jump over» some local maxima of the potential energy and so avoid some local minima of that energy. We present extensive experimental results on real 3-D medical images. In this particular application, we perform the matching process with the whole segmented volumes.