The wet delay, the excess radio path length due to atmospheric water vapor, has been derived from 71 days of microwave radiometer measurements at the Onsala Space Observatory, Onsala, Sweden. The temporal and spatial variability in the wet delay was analyzed. When we estimated daily variance rates, the parameter characterizing a random walk process, values in the range 3.1¿10-9 to 1.1¿10-7 m2/s were found for the timescales 10--20 min. We estimated horizontal gradients in the wet delay and found that the temporal variability of the gradient components changed significantly from day to day. The variations in the gradients were also found to be significantly larger if data acquired only at relatively high elevation angles were used in the calculation. The effect of the wet delay variations on Global Positioning System (GPS) geodetic estimates was analyzed by performing Monte Carlo simulations. We used a Kalman filter with parameters for geodetic GPS data processing, first modeling the atmosphere as a horizontally homogeneous random walk process in time. In this case the estimated wet delay was found to be more sensitive to a detuning of the Kalman filter than the vertical component estimates. The RMS errors in the wet delay estimates increased from 2.2 to 3.6 mm when the atmospheric variance rate changed from 1.0¿10-8 to 1.0¿10-7 m2/s and when the filter parameter was set to 1.0¿10-8 m2/s. When simulated wet delay gradients were added to the data, it was seen that if gradients are not estimated by the Kalman filter on days with large gradient variability, the scatter introduced by the gradients can dominate the other modeled error sources. ¿ 1998 American Geophysical Union