Two numerical experiments of ensemble hindcasting for 17 years are performed with five-member ensembles in each experiment,and the initial conditions are NCEP-DOE Reanalysis 2 data(adopted in real-time prediction)for experiment 1 and supplied by the system of three-dimensional variational assimilation(3D-VAR)for experiment 2.The model used here is a grid-point Atmospheric General Circulation Model with 9 vertical levels and a resolution of 2°in latitudinal and 2.5°in longitudinal direction,which is developed by the Institute of Atmospheric Physics,the Chinese Academy of Sciences(IAP9L2x2.5-AGCM).The technique of Correlation Analysis is applied.Comparison results show that(1)the difference of temporal anomaly coefficient between experiment 1 and 2 is very small in tropics,whereas it is large in middle and high latitudes,especially in East Asia(China),(2)temporal anomaly coefficient of most variables in experiment 2 is bigger than that in experiment 1 in most of China region,(3) spatial anomaly correlation of many variables have some improvement by use of the initial conditions after data assimilation(in experiment 2),due to the inclusion of atmospheric information for some time and the resulting coordination with dynamic model.The results demonstrate quantitatively that the role of the initial conditions on the shortterm climate prediction in East Asia is more important than in the global,what's more,the initial conditions after data assimilation are better than that used in real-time prediction to some extent.