Three months of continuous data from the Global Positioning System (GPS) using 20 sites in Sweden and 5 sites in Finland have been used to estimate the integrated amount of atmospheric water vapor. The quality of the data has been assessed by comparisons with a microwave radiometer (water vapor radiometer (WVR)) at the Onsala Space Observatory and with data from four different radiosonde stations. We found the agreement in integrated water vapor (IWV) between the GPS estimates and the radiometer data to be 1--2 kg/m2 in terms of daily root-mean-square (rms) differences. A major part of these rms differences were caused by a bias between the data sets. This bias (WVR--GPS) varied from day to day between -1.0 and +2.5 kg/m2 with a mean value of +1.3 kg/m2. Comparisons with radiosonde data showed rms differences around or slightly above 2 kg/m2 for each station using the entire 3 month data set. Also here the GPS estimates were, on the average, below the radiosonde results. We show that the radomes used to protect the GPS antennas are likely to cause a large part of the observed bias. Spatial structure functions were calculated by using the GPS and the radiosonde data. An overall consistency between the GPS-based and the radiosonde-based structure functions indicates that the spatial correlations between the GPS estimates are not affected by the estimation process used in the GPS data analysis. ¿ 1998 American Geophysical Union