Abstract: We screened for the highest performance model among several winter wheat yield predicting models. The selected model was weighted and integrated in order to improve the accuracy of prediction, as it plays a key role in ensuring food security. Daily meteorological observations, winter wheat yield data, and growth period observations were obtained from 69 basic meteorological stations in Jiangsu Province from 1993 to 2018. Then, five methods of meteorological yield and trend yield separation (linear separation, percentage difference, 5-or 3-year sliding average, and quadratic curve) were compared. On this base, by using the fitting test and hind-casting test, we evaluated and analyzed the simulation effects of yield prediction methods based on similar years with bumper or poor harvest, key meteorological factors and climate suitability, and integrated the methods for Jiangsu winter wheat yield prediction. The results revealed the following:1) For the same yield prediction method, the yield separation methods had a greater effect on prediction accuracy. The quadratic curve method was the best among the linear separation, percentage difference, 5-or 3-year sliding average and quadratic curve methods. The prediction accuracy of the weighting method was higher than the large probability method in the similar years with bumper or poor harvest prediction method. From 1993 to 2013, the average accuracy of the methods of the similar years with bumper or poor harvest prediction, key meteorological factor, and climate suitability were 89.67%, 94.86%, and 94.96%, respectively. 2) The accuracy of the integrated prediction method was more than 96.33% in the past 5 years, and it was higher than that of the similar years with bumper or poor harvest-weighting model, key factor-quadratic curve model and climate suitability-quadratic curve model. This could probably overcome the less stability of prediction accuracy of a single prediction method. 3) The closer the predicted time to the maturity period and the more comprehensive the prediction factor information, the higher the accuracy of the prediction model. These results provide a scientific basis for selecting an optimized prediction model for winter wheat yield in Jiangsu, and the methodology of model screening can also be used in other provinces.