Accurate estimations of grain output in the agriculturally important region of Northeast China are of great strategic significance for guaranteeing food security.New prediction models for maize and rice yields are built in this paper based on the spring North Atlantic Oscillation index and the Bering Sea ice cover index.The year-to-year increment is first forecasted and then the original yield value is obtained by adding the historical yield of the previous year.The multivariate linear prediction model of maize shows good predictive ability,with a low normalized root-mean-square error(NRMSE)of 13.9%,and the simulated yield accounts for 81%of the total variance of the observation.To improve the performance of the multivariate linear model,a combined forecasting model of rice is built by considering the weight of the predictors.The NRMSE of the model is 12.9%and the predicted rice yield explains 71%of the total variance.The corresponding cross-validation test and independent samples test further demonstrate the efficiency of the models.It is inferred that the statistical models established here by applying year-to-year increment approach could make rational prediction for the maize and rice yield in Northeast China before harvest.The present study may shed new light on yield prediction in advance by use of antecedent large-scale climate signals adequately. Accurate estimations of grain output in the agriculturally important region of Northeast China are of great strategic significance for guaranteeing food security. New prediction models for maize and rice yields are built in this paper based on the spring North Atlantic Oscillation index and the Bering Sea ice cover index.The year-to-year increment is first forecasted and then the original yield value is obtained by adding the historical yield of the previous year. The multivariate linear prediction model of maize shows good predictive ability, with a low normalized root-mean- square error (NRMSE) of 13.9%, and the simulated yield accounts for 81% of the total variance of the observation. To improve the performance of the multivariate linear model, a combined forecasting model of rice is built by considering the weight of the predictors The NRMSE of the model is 12.9% and the predicted rice yield explains 71% of the total variance.The corresponding cross-validation test and independent samples test further demonstrates the efficiency of the models.It is inferred that the statistical models established here by by applying year-to-year increment approach could make rational prediction for the maize and rice yield in Northeast China before harvest. The present study may shed new light on yield prediction in advance by use of antecedent large-scale climate signals adequately.