Considering the difficulties in star centroiding computation of star image that is polluted by strong noises, a centroiding algorithm based on threshold judging with sub-window gray distribution static evaluation was proposed. With certain number of sub-windows with certain size chosen based on the image character, the gray distribution was studied and the decentralization was calculated. After the judgment of decentralization according certain rules, sub-windows which contain peak noise regions or those unfavorable would be displaced by new chose ones. Hence, the threshold gray value final gained would represent the comprehensive character of the star image and the succeeding application of square-weighted subdivision algorithm could improve the succeeding centroiding algorithm’s accuracy. Simulation and result analysis proved a pleasant performance and robust capacity against sever peak noises of the proposed algorithm.