Abstract: Usual methods of small target detection couldn’t make satisfied result when process the sequence space images in which the feature of stars，targets and noise are similar and the stars have large gray range.The authors present a targets detection method based on the hierarchical clustering which constructs the similarity measuring function according the movement rule of stars and targets，finds the optimal separating hyperplane and classified amount by means of finding the inflexion of error square sum function，then distinguish the targets from stars and noises by two level hierarchical clustering.The results of experiments indicate that this method can deal with 8 bit image and 16 bit image and can detect the single point and multi-point targets efficiently from sequence star images.