The problem of landmark location is one of the key technologies of machine vision and pattern recognition. Active shape models(ASM) is a traditional method of landmark location, which has high degree of accuracy and robustness in some case. To improve the pupil landmarks accuracy of face image, we propose a method which combines Hough Transformation with ASM. The positions of the pupil are initialized by ASM algorithm, and Sobel operator is used in edge detection and Hough circle detection around the eye position. Through these steps, it can calculate the offset of initial position. The experiments on the face image which were captured in our laboratory room show this method can obviously improve the accuracy of the pupil landmarks. With the limit of Initialization the positions, computation is reduced with our control.