[Objective] To provide a reference for the flood-control safety of the loess plateau check dam system, a BP neural network combination model was tried to apply for predicting runoff from a storm-flood event.[Methods] The BP neural network(BPNN) combination model(BPNNC) was constructed on the base of multiple linear regression model(MLR) and detrended cross-correlation analysis(DCCA). Its output was compared with those from other three single models(MLR, BP neural network and DCCA) by the model evaluation indexes of mean square error(MSE), mean absolute error(MAE), mean absolute percentage error(MAPE), and deterministic coefficient(DC).[Results] The four values of MSE, MAE, MAPE and DC from BP neural network combination model were 2.144, 5.453, 0.074 and 0.988, respectively, which were better than the ones of the single models. The order of model precisions from high to low was BP neural network combination model, BP neural network model, multiple linear regression model and detrended cross-correlation analysis, successively.[Conclusion] The BP neural network combination model is more stable as compared with the single models, which can be used to predict the runoff from a storm-flood event.