Abstract: The character of high dynamic star sensor's sky image and the deficiency of existing star tracking algorithm at home and abroad were presented. Aiming at these deficiencies, a new star tracking algorithm based on Kalman prediction was put forward. The model of stars' movement was set up based on the character of the star sensor's movement. The adaptive Kalman filter was used to predict the position of the reference stars. The star was matched and tracked by Star Neighborhood Approach. At the end of the article, the prediction and tracking results were presented. The experiment results indicate that the star position prediction errors are less than 5 pixels under the dynamic condition of 5()/s, and the success rate of tracking is up to 95%. The method can adapt for high dynamic star sensor and improve the success rates of tracking availably.