On-orbit calibration is aimed at revising the star trackers' measurement model parameters and maintaining its attitude accuracy. The performance of existing calibration methods is quite poor. Among all the model parameters, the estimation of the principal point location is very challenging due to its vulnerability against measurement errors, yet, that it is the only parameter depicting the optical axis' projecting position on the image plane makes it of great significance. Its estimation error adds fixed bias to the output attitudes. Based on the criterion of vector pattern match, an on-orbit calibration method is proposed. The principal point location is estimated according to the criterion first. The other model parameters are updated by maximum likelihood method, and measures of multiple succeeding frames optimization and star density weight are adopted in the method to guarantee the estimation of robustness. Simulation and night sky observation results proved the validity of the proposed method. In the simulation with a poor initial guess of the principal point location, novel method's result is better than the least square method and Samaan's method.