Objective: In navigated orthopedic surgery, accurate registration of bones is of major interest. Usually, this registration is performed using landmarks positioned directly on the bone surface. These landmarks must be exposed during surgery. Our goal is to avoid the exposure of bone surface for the sole purpose of registration by using an intraoperative ultrasound device that can localize the bone through tissue.Method: We propose an algorithm for the registration of C T and ultrasound datasets that takes into account the fact that ultrasound produces very noisy images (speckle) and shows only parts of the bone surface. This part is made horn the C T dataset. Next, a surface volume registration is performed by searching for a position of the estimated surface that maximizes the average gray value of the voxels in the ultrasound dataset covered by the surface.Results: The algorithm was implemented and validated using an ex vivo preparation of a human lumbar spine with surrounding muscle tissue. On the basis of this data, the method has a large radius of convergence and a repeatability of 0.5 l l~m for displacement and 0.5 degrees for rotation.Conclusions: A robust algorithm for the registration of 3D C T and ultrasound datasets is presented. The computation time seems sufKciently short to permit intraoperative use. Comp Aid Surg 7:146-155 (2002). QZOOZ Wiley-Liss, Inc.