When fault occurs in different branches of radial distribution networks，voltage traveling waves will get to the measuring point installed on the bus side along paths of different branch combinations，leading to differences in natural frequency and its amplitude distribution of the transient voltage obtained from the measuring end.As there is a mapping relationship between the distribution and amplitudes of the natural frequencies and the propagation paths，we can use the strong nonlinear fitting capability of artificial neural networks（ANN）to realize the fault location and branch identification.By using the natural frequency of zero-sequence voltage data abstracted from a quarter of the industrial cycle after fault occurs as input samples of the layered and distributed ANN model to achieve fault location firstly.Employ the amplitudes corresponding to the natural frequencies as input samples to realize branch identification，outputting the fault distance and branch number of the failure point.Large numbers of electromagnetic transient simulations show that the method presented is effective.