Abstract: The planting areas of winter wheat in Shandong province were divided into four regions by the newly developed statistical method，cluster analysis of statistical test （CAST）.And models for forecasting winter wheat yield in each sub region were established and used to predict yield according to the climatic suitable index which based on temperature suitability，sunshine suitability，water suitability，that built by using winter wheat yield data，winter wheat growth data，daily meteorological data and moisture content data，from 1981 to 2011 Meanwhile，we compared it with the method of traditional contour partitions.The results indicated that，one region could not pass the significant test resulting in no prediction model with the method of contour line，and the average accuracy of others for historical forecasting was 94.2% and extrapolated forecasting was 92.3%.In relative terms，with the method of CAST，each region all passed the significant test and prediction models were established in each sub region，and the average accuracy for historical forecasting reached 95.8% and extrapolated forecasting reached 93.6%.The results showed that the yield prediction based on CAST was obviously superior to contour line and could provide an important way for fine yield prediction in agricultural meteorology.

Key words: Yield prediction, Traditional regionalization, CAST regionalization, Climatic suitable index, Winter wheat