The large regional-scale crop yield forecasting is of great significance to ensure national food security and agricultural sustainable development. To predict regional-scale winter wheat yield, the winter wheat forecasting method proposed by Becker-Reshef was implemented in major winter wheat planting areas in China. Firstly, the winter wheat planting areas were extracted through time series of MODIS NDVI with 250 m spatial resolution in Hebei, Henan and Shandong Provinces. Winter wheat adjusted NDVI peak at the heading stage was used to analyze the correlation with winter wheat yield on the purest wheat pixels from 2000 to 2010. Winter wheat yield prediction models were established through the regression statistical relationship between NDVI peak at heading stage and winter wheat yield at the prefecture-city unit from 2000 to 2009. The results showed that the dense wheat planted cities had good model accuracy. Finally, the accuracy of prediction models was evaluated by using the statistical winter wheat yield in 2010. The results showed that winter wheat yields could be forecasted two months in advance with average forecasting error of 7.49%, and the yield forecasting method using adjusted NDVI peak at heading stage had considerable potential applications at the prefecture-city scale in China. The study provides basis and a method for other crops yield forecasting in agricultural regions.