Multipath is the dominant error source for high-accuracy positioning systems since it is uncorrelated and uncertain at different location and can not be eliminated by the differential technology. To this end, it is significant for eliminating the multipath error and improving the positioning accuracy to estimate multipath parameters. In this paper, the multipath estimation problem is transformed into a parameter estimation problem, and the particle filter (PF) algorithm is applied for multipath estimation. However, there is a drawback of particle impoverishment in standard PF by using re-sampling strategy, which may lead to PF converging to a wrong value. In order to solve this problem, a differential evolution (DE) algorithm instead of the resampling strategy is used to generate new particles in PF. The proposed algorithm is named as improved differential evolution-based particle filter (DEPF). In DEPF, DE algorithm is used to promote the new particle moving towards the true post probability density distribution of the target state, which can avoid particle impoverishment. Simulation results show that the DEPF algorithm outperforms PF and EKF for multipath estimation in non-Gaussian noise.