In recent years, 3D point cloud registration has a very important role in medical diagnosis and surgical planning, and the accuracy of registration has always been the focus of attention. Several common registration algorithms and performs multiple sets of registration experiments on Bunny, Dragon and Hippo 3D point cloud models in third-party 3D model libraries are introduced. The registration accuracy is verified. Experimental results show that the Super-4pcs algorithm has different angles and low overlap rate for 3D point cloud model, and better registration results can be obtained comparing with several general registration algorithms. At a low overlap rate, a high precision registration effect of 0.001 m can still be achieved. It provides some theoretical guidance for doctors to more effectively apply the point cloud registration algorithm to medical diagnosis and surgical planning.