Abstract: The speed of star image registration affects the whole speed of the processing of the star image as star image registration is one of the most important steps of star image processing. In recent years, the general purpose computing of graphic process unit(GPU)has a rapid development. In this paper, the computing power of GPU for the general purpose computing and the problem of the speeding up of processing of star image registration were combined to study the accelerated processing algorithm based on GPU. A parallel model of GPU for the registration algorithm was proposed and CUDA programming language was uesd to realize it. Experiment result shows that the parallel model also fulfills the purpose of the image registration and has a 29.043X speedup compared with the serial CPU program.