In order to search for more timeliness and accuracy of forecasting methods, Yutian County, Hebei Province was chosen as the pilot site. A three-year field experiments were conducted on winter wheat. During different growing periods, three yield-contribution factors were measured in situ and the analysis was done. The results revealed that single-factor model of application allowed the prediction of time ahead of winter wheat yield to the heading stage; the prediction accuracy can reach to 88%. Prediction of two-factor model can be advanced to the heading stage to the early filling stage, and the AI is more than 90%. This study provided an effective method namely the multi-stage prediction model to effectively predict crop yield. It can overcome the regular operation problems and provide important references for agriculture warning system by remoter sensing at all levels of agricultural crop production forecast.