Rapid processing of large scale CCD star image is a key technique for new type of star sensor as well as space surveillance sensor. The foreground of CCD star image is differ form the background distinctly. Majority pixels of the image belong of the background. The typical histogram of CCD star image has a two-peak values structure. Base on those analyses, a strategy for star image＇s pretreatment was presented. The background and noise of the image is removed by a window transform of the pixel gray, all pixel whose gray value below the threshold will be set as a black pixel, the gray of which will be set as zero videlicet. The threshold is determined through the numeric statement of the background pixels by an iterative algorithm. Then a new algorithm for star point locating was presented by a two-step projection checking method. The horizon domains of the star points can be determined through the vertical projection checking of the image at first, then the horizon domain of each star point can be determined through the horizon projection in the horizon domains of the star points. The precise position of the star points is calculated using centroid method at last. The new algorithm is simpler and has smaller consumption of computing and memory than tradition connected domain methods. Experiments on real star images with size of 2048 × 2048 indicated the validity of the algorithm.