An autonomous predictive centroiding algorithm of unmatched stars is presented in this paper to obtain the real star centroidings of identified stars with star tracking algorithm. All the unmatched stars within Field of View(FOV) are searched from a guide star catalogue according to identified stars. Then, these ideal centroidings of unmatched stars are calculated through these identified stars. The algorithm can obtain the real centroidings of unmatched stars within some threshold scan windows of star image and can identify the centroidings of unmatched stars through the identified stars within FOV,so that recognition performance is improved. The algorithm is successfully demonstrated by real space experiments, experimental results show that the number of pixels of scanning star image by the proposed algorithm is only 0.12% of that by local obtaining star location algorithm,when unrecognized star locations are captured. The conclusions reported here prove that proposed method can improve the update rate of data.