In order to analyze the characteristics and applicability of the mathematical statistical model and the machine learning model in medium and long term runoff forecast, stepwise regression and random forest were selected to build a medium and long term forecast model. Based on the physical mechanism of meteorological factors, combined with single correlation coefficient and random forest importance analysis, key meteorological factors were identified and input into the model. The model was trained based on the inflow runoff of Wudongde and Pubugou reservoirs in the upper reaches of the Yangtze River from 1959 to 1998, and the inflow runoff of the two reservoirs from 1999 to 2014 was predicted. The results show that the two models have good training effect and strong stability. The prediction result of random forest is higher than that of stepwise regression, but the difference of accuracy is small. Random forest can reduce the fitting error caused by the abnormal change of predictor value, but the overfitting problem is more obvious. Keywords: reservoir inflow; medium and long term forecast; stepwise regression; random forest; meteorological factors; Wudongde Reservoir; Pubugou Reservoir; upper reaches of the Yangtze River 〖FL