The design of most ulnar head implants is currently not based on studies of anatomical variations. To enable prosthesis design based on anatomical features, a method is developed for the detection of the ulna surface that articulates with the radius head. The articulating surface is detected by combining partial articulating ulna surfaces detected in CT scans of different individuals in different poses. Correspondences between ulnae of different individuals are established through the construction of a statistical shape model (SSM) of the ulna head. The articulating surface is attached to this SSM, allowing the detection of articulating surfaces in ulnae that are not in the training set of the model. As a simple shape is desirable for prosthesis design, three quadratic surfaces were fitted to the articulating surfaces of 40 ulnae. The mean fitting error for the simplest surface, a cylinder, was 0.20 mm. As this error is smaller than the voxel size of 0.3 mm isotropic, it was concluded that the articulating ulna surface can be satisfactory approximated by a cylinder part.