Abstract: Ballistic midcourse is the main stage for ballistic missile defense and it is also the stage that has the most complex targets. Many heavy, light decoys and body debris in midcourse make it hard to get convincing recognition result based on single feature or one measurement. Based on the micro-motion and structure features of midcourse target, Basic Probability Assignment Function(BPAF) of each single feature is got via BP neural network firstly. Then, multi-feature fusion at current moment is done with D-S evidence theory. Finally, recognition results from different moments are fused with D-S evidence theory again to get the final results. Simulation results show the effectiveness of this method.