Most existing 3D ear recognition methods are based on the iterative closet point (ICP) algorithm that requires large amount of calculation and long matching time. It is easy to fall into the local optimum problem. Meanwhile, the ear region used for registration contains much redundant information. To solve these problems, a fast 3D ear recognition method is proposed in this study, which is based on the local and global information. Firstly, key points are detected according to internal shape features. In this way, the normalization of the ear region is achieved. Secondly, the lowdimensional local features are extracted to realize key points match. The candidate list is obtained, and the registration process is realized by fast point feature histogram and improved ICP algorithm with the normal vector of the point cloud. The proposed algorithm is evaluated by using the UNDJ2 database. Experimental results show that the key point feature extraction only takes 0026 s, and it takes 0015 s for key point match. The identification experiments show that Rankone recognition rate can reach 9855%, which shows that this method has faster recognition speed and higher recognition rate than other stateoftheart algorithms.