Precipitable water vapor amounts (PW) determined by Global Positioning System (GPS), radiosonde and operational numerical weather prediction (NWP) system analysis at three stations (Naqu, Gaize, and Deqin) on the Tibetan Plateau are compared. PW measured by water vapor radiometer at Naqu and a low-elevation site, Xian, is used for calibration. The results show that the PW determined by NWP analysis in these regions is comparable with that of the radiosonde measurements but that they both are systematically smaller than those determined by the GPS measurements. The averaged difference of PW between GPS and radiosonde estimates is about 1.75 mm, and that between GPS and NWP analysis can be as large as 7.75 mm. These differences are relatively larger than those reported in the literature because the absolute PW in this region is much smaller. The effect of such large differences on the surface radiation budget is evaluated using a radiation model. The results show that both longwave and shortwave radiative fluxes at the surface determined using the model analysis profiles with the water vapor corrected by the GPS PW are closer to the observations compared with those without water vapor correction. The flux difference at the surface with and without water vapor correction is about 20 W m-2 in the shortwave and 30 W m-2 in the longwave. These differences are much larger than that caused by doubling the concentration of carbon dioxide in the atmosphere in this region.