In order to solve the problem that the impact point prediction time is too long and the flight trajectory can not be predicted clearly and timely, an improved impact point prediction method is proposed. A reference model coordinate system is established based on the trajectory model, the parameters of the tracking equipment are configured in the reference model coordinate system, and the tracking equipment data collected in real time is filtered by an unscented Kalman filtering (UKF) algorithm to obtain a fusion trajectory. The Runge-Kutta algorithm is used to extrapolate and predict the impact point. Through Matlab simulation analysis and actual effect verification, the results of this method are more accurate and more applicable.