Abstract: BP neural network is used to improve real-time and robustness for star pattern recognition algorithm of star sensor. According to the trajectory, the navigation stars and corresponding pattern were reduced, that is, the star chart in FOV (field of view) corresponds to one and only navigation pattern. The largest empty circle on plane is computed to construct the exhaustive set in FOV by two-dimensional Voronoi diagram, and then the exhaustive set is constructed. By iterative comparisons, it is found that the navigation pattern of all the stars′ convex polygon in FOV can be composed if the fixed stars are chosen as the convexes, and taking its angle distance and vertex angle as the recognition vectors has the advantages of translation and rotational invariance. The simulation experiment shows that the success rate of accurate recognition is one hundred percent and the time of recognition is less than 20ms. Therefore, the recognition algorithm has a certain utility value.