This paper describes a novel method for image-based, minimally invasive registration of the femur, for application to computer-assisted unicompartmental knee arthroplasty (UKA). The method is adapted from the well-known iterative closest point (ICP) algorithm. By utilising an estimate of the hip centre on both the preoperative model and intraoperative patient anatomy, the proposed 'bounded' ICP algorithm robustly produces accurate varus-valgus and anterior-posterior femoral alignment with minimal distal access requirements. Similar to the original ICP implementation, the bounded ICP algorithm converges monotonically to the closest minimum, and the presented case includes a common method for global minimum identification. The bounded ICP method has shown to have exceptional resistance to noise during feature acquisition through simulations and in vitro plastic bone trials, where the its performance is compared to a standard form of the ICP algorithm.