In order to study the dynamic response of riverbed deformation due to the changes of the water and sediment conditions, this article introduces a method of BP neural network to establish mathematic method to predict the riverbed deformation from the two aspects of riverbank deformation and river cross section deformation. The model was trained and verified by the measured data of Shishou bend in Jingjiang River. The model was used to predict the riverbed deformation on section Jing-92 and section Jing-96 on year 2008 after the application of the Three Gorges Reservoir. The results can reflect the evolutionary tendency of riverbed deformation, which show that the model is feasible to be used to predict the riverbed deformation.