Spinal fusion is a high-risk surgical procedure and requires high accuracy to accomplish its surgical outcome. The current developed computer-assisted surgery has been proven effective for accuracy improvement. However the accuracy of computer-assisted surgery is strongly based on registration that constructs the relationships among patient, image model and the surgical instruments. A large error of registration will result in excessive surgical errors that could cause irreparable results. This study investigated the influence of the amount of sampling points from different regions of vertebral surface region and of the trajectories sampling the points to registration accuracy. The optimal numbers of sampling points on different regions of vertebral surface were observed and analyzed. The sampling trajectories which could produce better registration results were also experimentally verified.