A dynamical downscaling is performed to improve the regional climate 6 simulation in China. It consists of using a variable resolution model LMDZ4 nested 7 into three global climate models (GCMs): BCC-csm1-1-m, FGOALS-g2 and 8 IPSL-CM5A-MR, respectively. The regional climate from different simulations is 9 assessed in terms of surface air temperature and rainfalls through a comparison to 10 observations (both station data and gridded data). The comparison includes climatic 11 trends during the last 40 years, statistical distribution of sub-regional climate, and the 12 seasonal cycle. For surface air temperature, a significant part of the improvement 13 provided by LMDZ4 is related to the effect of surface elevation which is more 14 realistic in high-resolution simulations; the rest is related to changes in regional or 15 local atmospheric general circulation. All GCMs and the downscaling model LMDZ4 16 are, more or less, able to describe the spatial distribution of surface air temperature 17 and precipitation in China. LMDZ4 does show its superiority, compared to GCMs, in 18 depicting a good regional terrain including the Tibetan Plateau, the Sichuan Basin and 19 the Qilian Mountains.