Water vapor is one of the major components of the atmosphere and material resource for rainfall. The spatial distribution and temporal variation of precipitable water vapor on the Tibetan Plateau play an important role in the regional weather and climate. The reliability of precipitable water vapor measurements is greatly of concern. The characteristics of the systematic and random errors of the radiosonde (RS) precipitable water (PW) data by comparison with ground-based GPS measurements are studied at Lhasa during the period from year 1999 to 2010 and at Naqu in year 2003. The results show that the radiosonde PW is significantly drier than GPS PW at Lhasa during a period of more than one decade. Different types of radiosonde humidity sensors show different magnitudes of PW dry biases. GZZ-2 (goldbeater’s skin hygrometer) and GTS-1 (carbon hygristor) have relative mean dry biases of -8.8% and -3.9% and relative mean random errors of 19.8% and 13.6%, respectively. Due to the introducing of the high performance humidity sensors (GST-1), the relative PW difference is apparently reduced over the past 10 years. The temporal variation characteristics of the radiosonde PW dry bias are also investigated. The results show that the radiosonde PW dry bias exhibits pronounced diurnal and annual variations. The dry bias of the radiosonde PW is much larger at 1200 UCT than that at 0000 UTC, and larger in summer than that in winter. The relative mean PW biases for GZZ-2 and GTS-1 are respectively -15.8% and -7.3% at 1200 UTC, and -1.6% and -0.4% at 0000 UTC at Lhasa. The relative mean PW bias for GZZ-2 is -12.4% at 1200 UTC, and -0.3% at 0000 UTC at Naqu. Additionally, the causes of diurnal and annual variations of the radiosonde PW dry bias are analyzed. The solar radiative heating to the humidity sensors may have played an important role in the radiosonde PW dry bias diurnal and annual variations. It can be seen that the diurnal variations of the radiosonde PW dry bias are significant partly because the air temperature is higher at 1200 UTC than that at 0000 UTC. The annual variations of the radiosonde PW dry bias are pronounced partly because the air temperature is higher in summer than that in winter. The calibration methods for the radiosonde PW dry bias are developed and applied to the GZZ-2 and GTS-1 sounding PW datasets at Lhasa and Naqu. The corrections greatly improve the accuracy of the radiosonde PW.