With the method of BP neural network, simulation and prediction of runoff generation and sediment yield in four different runoff plots(farmland, wood land, artificial grassland, and abandoned land) are studied.Relative errors of runoff generation in four different plots are 0.2%- 5.7%, 0.1%- 2.5%, 0.7%-2.9%, and 0.1%-3%, respectively; relative errors of sediment yield, 0.1%-3.2%, 0.2% -3.1%,0.6%-4.2%, and 0.2%-2.7%; maximum relative errors of runoff generation, -11%, 14%,-14.6%, and 18%; the maximum relative errors of sediment yield, 10.9%, 27.3%, 15.0%, and 26.3%.The results show that the effect of simulation and prediction of runoff generation and sediment yield using the met hod of BP neural network is good and that application of this method to the analyses of impound and intercepting sediment from runoff plot is feasible.