This paper gives an approach on quantitatively forecasting catchment sediment yield with neural network BP model. Because soil, geography and physiognomy have a considerable stability above the cross-section at the Shibazi Station in the Wujiang River Basin in a period, catchment sediment concentration has been forecasted by using four essential factors of plant-recover-rate, yearly rainfall, yearly average flow and yearly run-off quantity. Optimized BP model not only has high simulation accuracy, but also gives a good forecasting effect. This has provided a new method for quantitative research and scientific proof for the project of stopping sediment measurement at the Shibazi Hydrometric Station.