Abstract. Some medical interventions require knowing the correspondence between an MRI/CT image and the actual position of the patient. Examples are in neurosurgery or radiotherapy, but also in video surgery (laparoscopy). Recently, computer vision techniques have been proposed to find this correspondence without any artificial markers. Following the pioneering work of [GLPI+94], [CZH+94], [CDT+92], [SHK94] and [STAL94], we propose in this paper an alternative approach. We propose to trade the laser range finder for two cameras. Hence, we get dense reconstruction of the patient's surface and this allows us to compute the normals to the surface. We present a new method for rigid registration when surfaces are described by points and normals. It does not depend on the initial positions of the surfaces, deals with occlusion in a strict way and takes advantage of the normal information. Results are presented on real images.