Cone-beam computed tomography (CBCT) is widely used to get high spatial resolution images. However, a small amount of involuntary patient motion can degrade the quality of CBCT images, resulting in motion artifact. To address the issue, investigators have proposed various motion compensation algorithms. In this study, we present a novel method to compensate the patient motion using registration of opposite projections. The novel method estimates a translating rigid motion of a patient by finding the motion vector which maximizes the two dimensional correlation coecients of the opposite projections. The estimated motion is compensated by shifting a back-projection source in a reconstruction process. Results of real data and simulations show that the proposed method successfully compensates the motion artifact.