Osteoarthritis are one of diseases that people cannot avoid with aging. Moreover, some young people suffer from osteoarthritis because of sports and labor that give large load to joints. It is very important for the diagnosis and treatment to understand three-dimensional movement information on the joint quantitatively. In this study, the three-dimensional positional posture presumption technique of the bone was carried out from X-ray image. We focused on the elbow joint because the disease of the elbow joint is observed as frequently as knee joint. We performed 3D kinematics measurement at elbow joint from template pattern matching. In this study, 50 samples were estimated to examine the relation between length of outline and error margin. The results showed that higher accuracy was obtained by using longer outline information. In order to achieve higher accuracy, the feature vector selection method was adopted. The combination of template pattern matching and feature vector selection method improved the estimation accuracy from (θx,θy)=(1.44,2.08) to (1.26,1.72).