Based on the analysis and identification of the annual runoff sequence components of the Danjiangkou Reservoir, deterministic components such as the trend term, the jumping term and the periodic term were derived by using linear trend regression analysis method, sequential cluster method and variance spectrum method, etc. A stochastic auto-regression model of annual runoff based on Ensemble Empirical Mode Decomposition(EEMD)was proposed(EEMD-AR)and it was applied to the stochastic simulation and prediction of the annual runoff in the Danjiangkou Reservoir. Through the EEMD decomposition, the problem that stochastic simulation and prediction by auto-regression(AR)model cannot be directly applied due to the non-stationary historical runoff sequence of the Danjiangkou Reservoir has been solved. The simulation results show that EEMD-AR model can simulate and predict the annual runoff sequence of the Danjiangkou Reservoir in a good forecast accuracy and it maintain the statistical characteristics of the original historical sequence.