The non-linear and non-stationary characteristics of runoff series make it difficult to forecast runoff with high accuracy. In this paper, the EEMD-GRNN coupled model is formed by combining the EEMD and GRNN models, and the runoff series is decomposed into deterministic and stochastic components and predicted separately by the GRNN model. The sum of the predicted values constitutes the final predicted value of runoff. The EEMD-GRNN coupling model is applied to the middle and upper reaches of Yuanjiang River and compared with the prediction effects of other models. The results show that the EEMD-GRNN coupled model has a higher prediction accuracy and is effective in predicting the overall trend of runoff, but the stochasticity simulation needs to be further improved. The EEMD-GRNN coupled model performances better than BP, GRNN and EEMD-BP models, and can effectively improve the accuracy of runoff prediction and provide decision support for water resources optimization in the basin.