When the number of observed stars is few, the matching number of stars and the decision conditions of boundary distance do not satisfy the requirement the grid algorithm. Meanwhile because of the principle limits of the grid algorithm, the identification rate decreases obviously. According to these problems, an improved grid algorithm is put forward, and the identification rate of observed star increases effectively in precondition of ensuring the robustness. Experimental results show that the improved grid algorithm urges the identification rate increased from 95% to 99% compared with the traditional grid algorithm, under the condition that the number of observed star is less than 10. At the same time, the non-recognition and false-recognition stars in the star chart have been carried on the analysis, it can help to promote grid algorithm for star identification further.