State estimation of distribution network is an important part of distribution management system. The data which was used for state estimation usually has random noise interference of different degrees and can"t be used for the operation analysis of distribution network directly. In order to obtain more accurate state information of distribution network, the measured data must be filtered. To settle the question that the flexibility of unscented Kalman filter (UKF)is poor, and the filtering accuracy is restricted by parameters and initial filtering values easily, and the importance density function selected by the standard particle filer(PF) is unreasonable, unscented particle filter(UPF)algorithm is applied to the State estimation of distribution network in this article. The algorithm combines UKF and PF,and combines UKF with the latest measurement information to generate the importance density function for PF. It transfers the particles falling in the prior probability density region to the high likelihood region. Then, the filtering performance of PF is improved. The results of IEEE33 node system show that UPF has better performance and flexibility than UKF and PF, and is an effective state estimation method.