Based on the new global surface temperature data set CMST (China merged surface temperature), we systematically evaluated the historical climate simulation results of the 27 global climate models which are participating in the International Coupling Model Comparison Plan Phase 5 (CMIP5) from 1900 to 2017. Historic model simulations for the period of 1900-2005 and projections for the 3 different representative concentration pathways (RCP2.6, RCP4.5, RCP8.5) for 2006-2017. According to the comparison of various parameters in Taylor diagram, 9 models with better simulation results (MT9) are selected to present the models ensemble simulations. Compared with the ensemble average of mean all models (MAM), under any of the three RCPS, MT9 is significantly closer to the observed value (MAM overestimates the temperature change) in both spatial distribution and time variation . On this basis, this paper uses the prediction results of MT9 to analyze the amplitude of Asia surface air temperature rises during 2018-2099. The result showed that MT9 is expected to have smaller temperature warming than those of MAM under all the RCPs in the Asian region, which is about 0.08℃ for RCP 2.6, 1.20℃ for RCP4.5 and 3.54℃ for RCP8.5. These results are all smaller than those from the MAM ensembles, thus are more reasonable. In addition, this paper also analyzed the spatial distribution of the surface temperature anomalies in Asia based on the MT9.