Abstract: The least predicting error sum of square(PRESS) and Kalmanfilter were applied to build predicting model in this paper. Firstly, we constructed a model on the basis of PRESS. Then, the prediction model was amended by Kalmanfilter. The predicted results showed that more forecasting errors of PRESS model were reducedd than by stepwise regression analysis, and the filtered forecasting distance errors of tropical cyclone weree less than unfiltered forcasting (less than 200km). So we suggested this prediction method is good for tropical cyclone tracks prediction.