Based on three RCP scenarios (RCP2.6,RCP4.5,RCP8.5),a statistical downscaling model (SDSM) was established by using observed meteorological data,ERA - 40 reanalysis data and 5 preferred GCMs output selected from 23 GCMs of CMIP5. Then,the climate change scenarios were predicted,including daily precipitation,highest and lowest ambient temperatures during 20212050 in the Heihe River basin,where is the second largest inland river basin in northwest China. The results show that the SDSM has a good predicting capacity for the ambient temperature in the basin. During calibration and validation periods,the coefficient of determination (R~2) and the coefficient of Nash - Sutcliffe efficiency coefficient (NSE) are both larger than 0.9,while the root mean square error (RMSE) is less than 20%. However,the SDSM show relatively lower simulation efficiency for precipitation with R~2 and NSE values of above 0.5 in most meteorological stations,except the stations located in the downstream desert areas. Compared with the baseline period (19762005),the annual mean precipitation simulated by different GCMs during 20212050 show a decline globally in one RCP scenario only. In the rest RCP scenarios,however,the precipitation fluctuates in a range of - 10% - + 10%. Specially,the precipitation depends on season and month largely,and it was more in summer but less in spring in most RCP scenarios. Note that the highest and lowest ambient temperatures exhibit a similar increasing tendency during 20212050 under all RCP scenarios. The increment of the highest ambient temperature is lower than the increment of the lowest ambient temperature,especially, both increments rise with increasing concentration of RCP.