End points, corner points, branch points and crossing points are important feature points of skeleton shape. A new approach to detect these feature points is proposed. First, a method of the fast lookup table is taken to detect prepared feature points which include both of real and unreal feature points. Then, by clustering these feature points and analyzing local structure, real feature points are obtained and unreal feature points are eliminated. The types of true feature points are identified and the quantative description of corner points is roughly given. The experiment results indicate that this method is easy of use and has high calculation speed. It plays an important role for some detection problems and character recognition.