The crop growth model has distinct advantages of clear mechanism and dynamic serial simulation, and it has been widely used in regional crop production forecasting. The research focused on the major problem that the uncertainty of yield prediction with crop model can not be quantized and the issue that the realtime impact of meteorological driving factors on crop model was not timely reflected. Simulations in the main wheat production areas in Baoding and Hengshui of Hebei Province were conducted and a method was proposed to build a historical meteorological dataset as weather inputs to drive the WOFOST model to simulate wheat growth during the forecasting period. Then the yield was continuously forecasted through realtime updating. In this way, the traditional singlevalue forecasting of yield was changed to an ensemblebased probabilistic forecasting, and the regional yield ensemble forecast and event possibility forecast can be generated in everyday in the growing season. The validation results indicated that the WOFOST model with historical meteorological data was able to reflect the uncertainty of regional weather data. In the regional ensemble forecasting of wheat yield, the highest yield forecasting accuracy was achieved in the period from heading to grain filling stage. The correlation coefficient (PCC) between the ensemble mean and the measured yields was 0563, while the minimum average absolute error (MAE) was 458kg/hm2, however, the improvement of yield forecasting accuracy along with forecasting date was slow, because simulation with homogeneous input parameters in potential level was a little coarse. It suggested that with the help of remote sensing data assimilation or medium weather numeric forecasting, yield prediction could achieve better accuracy. The results showed that the regional yield ensemble forecast had strong feasibility, and this research provided a reference for the application of numerical weather forecast and quantification of the uncertainty in crop simulation system.