Based on the TIGGE datasets including the European Centre for Medium-Range Weather Forecasts(ECMWF), the U.S.National Centers for Environmental Prediction(NCEP), the United Kingdom Met Office(UKMO) and its multi-center ensemble systems, Bayesian Model Averaging(BMA) probabilistic forecasts of winter surface air temperature over East Asia are established.Anomaly correlation coefficient(ACC) and root mean square(RMSE) are used for the evaluation of the BMA deterministic forecasts.Furthermore, Brier score(BS), Ranked probability score(RPS), BSS and RPSS are applied to evaluate the performance of BMA probabilistic forecasts.The results show that the BMA forecast distributions are considerably better calibrated than the raw ensemble forecasts, and BMA forecasts of ECMWF, NCEP and UKMO EPSs provide better deterministic forecasts than the individual model forecasts.The BMA models for multi-center EPSs outperform those for single-center EPS for lead times of 240-360 h, and the optimal length of the training period is about 35 days.In addition, BMA provides a more reasonable probability distribution, which depicts the quantitative uncertainty of the forecasts.The uncertainty on the land(higher latitude) is larger than that on the sea(lower latitude).