The predictors are the surface temperature over the Eurasian continent,500hPa heights over the Northern Hemisphere,SST(sea surface temperature)over the tropical Indian Ocean and SST over the North Pacific.The canonical correlation analysis(CCA) prediction is used to establish forecasting relation.Then the ensemble canonical correlation prediction(ECC) method is used to predict winter temperature over China.The forecast skill is analyzed and independent samples test is also performed.The results show that different predictors have different forecast skills in various regions.Using the sea surface temperature over the North Pacific as a predictor,the skill is higher.But the ECC gives better performance,and its skill is higher than any CCA forecast of single predictor.Under the condition of the same predictors the regression ensemble forecast is better than the simple equal ensemble forecast.