In this article, the problem of extracting star spots for star sensor in stray light is considered. Under this setting, the effect of extracting star spots becomes worse, which affects the accuracy and reliability of the estimated attitude. An extraction method based on optimal surface fitting is proposed for this task. First, the imaging characteristics of stray light in star sensor are analyzed, and the surface model with a closed-form solution is built. Then, the method of estimating optimal stray light background and extracting star spots is proposed. The performance of the proposed method is verified by the simulated star images and star sensor images. Experimental results show that the detection rate, false detection rate and centroid accuracy obtained by the proposed method are better than that obtained by the methods based on threshold and morphological, which shows that the proposed method can resist the disturbance of stray light.