In order to analyze the influence of initial atmospheric conditions on short-term climate numerical prediction,two different sets of ensemble hindcasting experiments were performed with IAP9L2x2.5-AGCM(a grid-point Atmospheric General Circulation Model which was developed by the Institute of Atmospheric Physics,the Chinese Academy of Sciences) forced by observed and climatic surface sea temperature(SST),sea ice respectively for seventeen years(1988—2004).In each set of the experiment,there were three tests under three different initial atmospheric conditions,namely,NCEP-Ⅱdata,smoothed 5-day NCEP-Ⅱdata and smoothed 11-day NCEP-Ⅱdata.The method of correlation analysis was used here.The results showed that the influence of initial atmospheric conditions on short-term climate prediction in the extratropical areas is greater than that in the tropical areas.As to summer in East Asia(China),the initial atmospheric conditions of smoothed 5-day NCEP-Ⅱ data performed better than smoothed 11-day NCEP-Ⅱdata and NCEP-Ⅱdata(adopted by real-time prediction at present) by comparing the temporal anomaly correlation and the average of spatial anomaly correlation of meteorological elements during the 17 years.In contrast to the initial atmospheric conditions adopted at present,the work in this paper unveiled some useful information that is suitable for short term prediction for IAP 9L2°×2.5°-AGCM,which is valuable.