Operational prediction works, experimental predictions and predictability studies on short-term(monthly to interannual scale) climate prediction during the last ten years or so were reviewed. It is indicated that correlation coefficients(C.C.) between predicted and observed monthly mean 500 hPa hight anomalies vary between 0. 2 and 0. 3 independently what model or statistical method was used. Temperature prediction shows similar skill, but it is much lower for precipitation prediction. Seasonal prediction was mainly based on statistical method. CGCM used in seasonal prediction is only being in the first stage of the development. Seasonal rainfall forecasting in China showed nearly the same skill(0.2～0.3 of C. C. ). Central task of seasonal prediction is of improving the ability to predict the changes in SST, ice-snow cover and land surface conditions. The most successful skill was found in ENSO predictions, C. C. was above 0. 6 with six or more months lead. But, it is still suffered from the predictability barrier in spring. Therefore, operational short-term climate predictions in the next a few years should be as usual based on the statistics rather than on the dynamics. However, great efforts should be made in studying the dynamic and thermodynamic mechanism of the climate system, and improving the GCM. Without these efforts, it is impossible to increase significantly the prediction skill.