Monthly mean sea surface temperature (SST) is characterized by non-stationary and nonlinear feature. It is obviously unreasonable to apply linear data processing methods directly to non-stationary and nonlinear time series, which would produce large prediction errors. In order to improve the prediction accuracy and better address the non-stationary and nonlinear sequence prediction problem, in this paper, we present an example based on monthly mean SST anomalies (SSTA) of the Northeast Pacific (40°N-50°N, 150°W-135°W). We first use ensemble empirical mode decompose (EEMD) and complementary ensemble empirical mode decomposition (CEEMD) to decompose monthly mean SST into a series of Intrinsic Mode Function (IMF). BP (Back Propagation) neural network model is then utilized to predict each IMF. Finally, the forecast results of each IMF are reconstructed to obtain the predicted value of monthly mean SST. Results of the experiment indicate that the accuracy of CEEMD is better than that of EEMD, and CEEMD has improved the forecast accuracy based on BP neural network. Statistical analysis of the results of a series of experiments shows that this method is effective for SST prediction at the 1-year scale.