A star pattern recognition method based on subdivisiom feature set was proposed in order to quickly and efficiently recognize star pattern and accurately complete celestial navigation task. Firstly, a database was built by star data, on which feature star database was established by triangulation. Then subdivision feature of star pattern could be compared, which will be recognized, with feature star database and implement star pattern recognition. By improving some similar methods like Hamming similarity and Euclid similarity, a new star pattern recognition method was put forward based on triangulation feature set, by which a very small possible star set could be found. This process can be repeated to obtain adjacent possible star set. In these two star sets, the nearest star between right ascension and declination was the one recognized. Experiments show that accuracy rate can reach more than 97% by using this method, and the star pattern recognition task can be completed accurately.