A model-based method is described which defines a rigid transformation between two co-ordinate systems that reduces the accuracy requirements on the quality of the data-set (including, but not restricted to, the error in the acquisition process, and the number and spread of the points) measured in one of the two co-ordinate systems by identifying a set of remote correspondences that are used to bind the convergence process. The method can be used in minimal-access orthopaedic surgery to improve the accuracy of limb registration. Specific instances include femoral registration, by estimating the functional centre of the hip joint in both co-ordinate systems to be co-registered, and tibial registration, using the ankle centre as a distant set of paired correspondences. Accuracy can be measured in a variety of ways, including, but not restricted to, evaluating the mis-alignment between the two co-registered objects.