Abstract: Based on historical simulation experiments and future projections under the RCP4.5 scenario of 35 climate models in the Coupled Model Intercomparison Project Phase five (CMIP5), the Bayesian Model Averaging (BMA) is applied to project the surface air temperature change and uncertainty over China in the 21 century. Results show a consistent increase in surface air temperature in China, and the air temperature change is larger in the winter than in the summer and larger in the north than in the south. It is very likely (>80%) that the surface air temperature will increase by 0.7℃ in northern China, and the probability of the same increase in southern China exceeds 50% at the early 21 century (2016-2035). The probabilities of the air temperature increase of 1.5℃ in northern and southern China in the middle 21 century (2046-2065) are larger than 80% and 50% respectively, and the same probabilities of air temperature increase of 2℃ at the end of 21 century (2081-2100) are found in northern and southern China. Meanwhile, characteristics of the estimated uncertainty are similar to that of the air temperature change, with the lowest uncertainty of less than 0.6℃ being found in the Tarim Basin, the southern Tibetan Plateau and southeast China. Large uncertainties of greater than 1℃ are found in northern Xinjiang, northern Northeast China Plain and southeastern Tibetan Plateau, where the reliability of the estimated results is the poorest. Besides, compared with that in the summer, the reliability of projected air temperature change in the winter is lower in most regions except the eastern Tibetan Plateau; and the reliability of the projection at the end of 21 century is higher than that at early 21 century, but with the same spatial distribution of air temperature change.