In order to solve the centroiding problem of an over exposed star in star image processing, a universal centroiding algorithm was proposed with “first local calculation and then comprehensive integration”. First, the existence of the over exposure stars was confirmed with real star images. Then, by analyzing the grey distribution of over exposed stars, the over-exposure scale of every row/column was detected according to the corresponding grey value and the square weighted mass centroiding was applied to calculation of primary local results. Finally,an over-exposure scale related to scaling scheme was used again to decide corresponding weights of local results and the final result was obtained after a last integration. The simulation results prove that in an all-random experiment mode the final centroiding accuracy has be better than 0.05 pixel even the random noise is 30 pixel. The proposed algorithm is validated to be competent and could satisfy the demand of the centroiding algorithm for the high-accuracy and good adaptability in real applications.