Abstract: As a satellite attitude measurement device, the calibration accuracy of the star sensor's principal point and focal length serves as the main factor affecting the accuracy of its attitude output during the on-orbit service procession. If the calibration process contains the star image point with large random measurement noise, this will lead to large deviation of the calibration result of the star sensor's main point and focal length. Aiming at this problem, a weighted on-orbit calibration method of principal point and focal length for star sensor was proposed. Firstly, the method established the on-orbit calibration model of the star sensor.Then we added a reasonable calibration weight to the least squares estimation. In this way, we found and removed the points with large random measurement noise. Finally, the weighted least-squares estimation result was served as the measurement, and the extended Kalman filter was used to process the star image. The simulation results show that, if there is a large error in the position of the star point, this method can remove the bad points where the random measurement noise is too large. The statistical deviation of the angular distance is about 1/10 of the traditional method. Compared with the true value, the accuracy of the calibration parameters is 0.219 9 pixel, 0.148 7 pixel and 3.38 μm, respectively.