Regional Climate Model (RegCM) simulations, which is based on the fundamental physical and metrological mechanisms, could produce high-resolution climate dataset, without being limited by the availability of meteorological stations in Central Asia. In order to analyze the climate change in Xinjiang from 1958 to 2001,this study used the RegCM model to downscale the ERA40 and NCEP/NCAR reanalysis datasets to a 40- km resolution, and compared the simulation results with three widely used extrapolation datasets (CRU, WM and APHRO). All datasets indicated increased temperature in Xinjiang, increased precipitation in southern Xinjiang, and decreased precipitation in the Tianshan mountainous areas. All datasets except for the APHRO interpolated data indicated increased precipitation in northern Xinjiang. Our analysis showed similar spatial patterns in temperature and precipitation between the RegCM simulated and the extrapolation datasets. However, the simulated regional mean temperature was -3℃ lower than that of the extrapolated datasets. The simulated precipitation in the Tianshan mountainous areas was about twice the value of the extrapolated datasets. Because 73% of the meteorological stations in the study located at the low-mountain areas or the desert plains were in relatively hot and dry climate regimes, extrapolation datasets based on observations at these stations tended to overestimate temperature and underestimate precipitation. Compared with the extropolation method, regional climate modelling considers the detailed topography/landsurface characteristics in the study region at meso-/micro- scales, and is capable to reflect the spatial variation of major climatological elements with higher precision. However, due to limited sparse field observations and a lack of high- resolution remote-sensing-based climate datasets, we are currently unable to fully evaluate the accuracy of the model simulated climate datasets in central Asia, especially in the mountainous areas.