Image registration is a technique usually used for aligning two different images taken at different times and/or from different viewing points. A key challenge for medical image registration is to minimise computation time with a small alignment error in order to realise computer-assisted surgery. In this paper, we present the design and implementation of a parallel two-dimensional/three-dimensional (2-D/3-D) image registration method for computer-assisted surgery. Our method exploits data parallelism and speculative parallelism, aiming at making computation time short enough to carry out registration tasks during surgery. Our experiments show that exploiting both parallelisms reduces computation time on a cluster of 64 PCs from a few tens of minutes to less than a few tens of seconds, a clinically compatible time.