Space survey camera image stabilization system demands high control accuracy and puts forward higher requirements on the guide star sensor. In order to improve the accuracy and bandwidth of guide star sensor, star spot location method combining prediction windowing and Kalman filter is presented. Using gyros to measure three-axis angular rate information, and establishing a prediction equation for the rough position of the star spot, the rough position of star spot is obtained. Windowing in a small neighborhood of the predicted spot on the CMOS detector can improve the operation speed. Using Kalman filter algorithm to correct the position of the star spot, the accurate position of star spot is finally obtained. Simulation results show that compared with traditional centroid method, image processing time per frame is reduced from 59 ms to 27 ms, and the standard deviation of star location result is reduced from 0.1 pixel to 0.04 pixel. The proposed method is an effective way to improve the speed and accuracy of the star spot location, providing certain value for the development of China′s space survey camera.