To improve the robust performance for the recognition of star map, in this paper, a neural network algorithm for the recognition of star map is proposed. And the full coverage sample classification set for target sky area is established. By means of the proposed method, not only the robust performance for the angular distance with error is improved, but also the speed and the accuracy of recognition is substantially improved. The simulation results show that the proposed algorithm can remain the accuracy of 99%, which the error of the angular distance is ± 0.02° ∼ ° 0.06° °