Based on the complex nonlinear characteristics of rainfall-runoff on slope

a three-layer feed-forward back-propagation network model for runoff in different landuses(grass and shrubby slope

cutting slope and plowing slope) is established.Structure of the model has five input variables

including vegetation coverage

rainfall intensity

gradient

antecedent soil moisture content

and soil bulk density and one output variable of runoff amount from single rainfall event.The network model is validated by using the data observed from field simulated rainfall experiment.The BP network model is compared with conventional method as well.Results show that the BP network model may improve the precision of forecast.