Registration of medical images has become an important area of research. In particular, registration of computed tomography (CT) lung images is of great interest to radiation oncologists planning radiation treatment for patients with lung cancer. A review of existing image registration methods, as well as preliminary numerical results, indicate that methods based on a constant pixel intensity assumption, such as traditional optical flow methods, cannot be expected to produce accurate registration of lung CT images. Nonlinear methods allowing variations in pixel intensities, though more costly than linear methods, promise to be more accurate for this application.