This paper presents a method for estimating precipitable water (PW) from GPS-derived zenith wet delay (ZWD) using numerical weather forecasts. The key idea is that ZWD, which is the time delay of GPS signals caused by atmospheric water vapor, is nearly proportional to PW. By multiplying ZWD by a factor (Π), which depends on atmospheric refractivity and temperature, PW can be estimated. However, the accuracy of this transformation depends on the mean temperature (Tm) of the atmosphere, which must be estimated a priori. The relative error in Π closely approximates the relative error in Tm, which can be predicted with an rms relative error of less than 1% using numerical weather models. The paper discusses the theoretical basis for this method, including the derivation of Π from atmospheric refractivity constants and the error budget associated with the transformation. It also addresses the uncertainty in the refractivity constants and the potential impact of errors in these constants on the accuracy of PW estimates. The authors show that the values of the refractivity constants can be determined from experimental measurements, and that the uncertainties in these constants are relatively small. The paper also presents a comparison between the predicted Tm using numerical weather models and the directly measured Tm from radiosonde launches. The results show that the predicted Tm has a relatively small error, which is important for accurately estimating PW from GPS data. The authors conclude that it is now possible to estimate PW from GPS data with an rms error of less than 2 mm plus 1% of the PW, and that the greatest potential improvement in estimating PW from GPS observations lies in developing better techniques for estimating ZWD. The paper also notes that the assumption that wet delay is entirely due to water vapor may not always hold, and that further research is needed to determine the frequency with which this assumption breaks down.This paper presents a method for estimating precipitable water (PW) from GPS-derived zenith wet delay (ZWD) using numerical weather forecasts. The key idea is that ZWD, which is the time delay of GPS signals caused by atmospheric water vapor, is nearly proportional to PW. By multiplying ZWD by a factor (Π), which depends on atmospheric refractivity and temperature, PW can be estimated. However, the accuracy of this transformation depends on the mean temperature (Tm) of the atmosphere, which must be estimated a priori. The relative error in Π closely approximates the relative error in Tm, which can be predicted with an rms relative error of less than 1% using numerical weather models. The paper discusses the theoretical basis for this method, including the derivation of Π from atmospheric refractivity constants and the error budget associated with the transformation. It also addresses the uncertainty in the refractivity constants and the potential impact of errors in these constants on the accuracy of PW estimates. The authors show that the values of the refractivity constants can be determined from experimental measurements, and that the uncertainties in these constants are relatively small. The paper also presents a comparison between the predicted Tm using numerical weather models and the directly measured Tm from radiosonde launches. The results show that the predicted Tm has a relatively small error, which is important for accurately estimating PW from GPS data. The authors conclude that it is now possible to estimate PW from GPS data with an rms error of less than 2 mm plus 1% of the PW, and that the greatest potential improvement in estimating PW from GPS observations lies in developing better techniques for estimating ZWD. The paper also notes that the assumption that wet delay is entirely due to water vapor may not always hold, and that further research is needed to determine the frequency with which this assumption breaks down.