The arid region of Northwest China is one of the regions with strong summer surface sensible heat flux transfer in Eurasia. Surface sensible heat flux over this arid region plays an important role in the variation of the East Asian monsoon system. However, large uncertainties in summer sensible heat flux over this region have been found in various reanalysis datasets, which hinders our exact understanding of its climatic influence on the East Asian monsoon system. Using the summer measurements from 2001 to 2014 at Dunhuang Gobi site, this paper evaluated summer sensible heat fluxes in four reanalysis datasets including NCEP/NCAR, NCEP/DOE, ERA-Interim, and JRA-55. Sensible heat flux observations at Dunhuang Gobi site were derived from routine meteorological observations combined with a reasonable surface flux parameterization scheme (Y08 scheme). The results indicate that the average summer sensible heat flux at this site is about 85.7 W m^-2, while its variation range is large due to the influence of local rainfall. There exist certain differences in the sensible heat flux among different reanalysis datasets. Compared with the observations, the sensible heat flux from the ERA-Interim is the best in both the magnitude and variation, and agrees well with observations when there is no local rainfall influence. The reasons for the uncertainty in sensible heat flux in the four reanalysis products and the differences between reanalysis and observations were investigated. Surface wind speed and differences between surface skin temperature and air temperature are associated with the specification of roughness and surface flux parameterization scheme. The difference between surface skin temperature and air temperature in the reanalysis products was much smaller than the observation (6.5 ℃) mainly due to the unrealistically high value of roughness and inappropriate parameterization scheme for the Gobi surface in these reanalysis products. The parameterization scheme of ERA-Interim is relatively better than those of others in that the difference between surface skin temperature and air temperature is close to observations, which results in a relatively reasonable estimation of sensible heat flux.