Abstract:  A series problems exist when tracking multi－targets in radar systems.The target loss and tracking error which are caused by the cross－overlapping of the targets trajectory still can’t be solved properly by now.In this paper，the weighted re－sampling method under union state is introduced which is based on the standard particle filter algorithm to track targets for the purpose of reducing target loss and tracking error mentioned above.The basic particle filter algorithm is optimized to simulate the posteriori probability of the targets by using discrete random variables.When there are multiple targets，the concept of joint states is firstly introduced in the paper.The state variables and observed variables are expressed by the correlation function defined above.Then the weights of particles will be updated by the state variables which has been sampled under union state by correlation function.So that the new particles can not only predict the posterior distribution of each target accurately according to the state change between targets，but also avoid unreasonable distribution of particles in the next time.The combined weighted－resampling algorithm can correct predictive value according the target motion.The target cross－motion does not affect the prediction and updating of the particle，and overcomes the problem of unstable tracking or large error by using this algorithm.The simulation results show that the improved particle filter can achieve the goal of tracking multiple targets correctly and the high accuracy of tracking multiple targets can still be maintained when the target trajectory crosses.