Abstract: The distillation of near-infrared star image of star tracker was analyzed in this paper. A new method for this problem was presented. Firstly, characters of near-infrared star image in wavelet domain were analyzed and mask used to detect stars in multi scales were designed according to the distribution of the gray of stars. Then, local threshold was applied to overcome the imbalance of the background under the constant false alarm rate. Lastly, positions of the heaviest gray values of stars in the secondly decomposed approximation coefficients were accessed, eliminating the effect of the single point noise and stripe noise. Experimental results indicate that the method could efficiently obtain the position of the heaviest gray value of detected stars with error around one pixel, which satisfies the requirement of estimation of centroid.