In order to improve the calculation accuracy and speed of distribution networks state estimation, this paper presents a robust algorithm of three-phase state estimation for distribution networks based on residual prediction. All kinds of measurements in distribution networks are used in the algorithm, Jacobian matrix becomes constant through phase transform of node voltage and branch current and equivalent transform of current amplitude measurement, and the amount of calculation is reduced due to Jacobian matrix remaining unchanged in the iterative process. Setting of equivalent weight based on prediction residual avoids the unnecessary iterations in the correction of the equivalent weight, which reduces the amount of computation. Simulation results show that the proposed method is stable and robust.