Microwave radiometer data and radiosonde data from the time period 1981--1995 have been used to study long-term trends in the integrated precipitable water vapor (IPWV). The two instruments have operated 37 km apart on the Swedish west coast. Model parameters are estimated for the entire data sets as well as for subsets of the data. The IPWV model parameters are a mean value, a linear drift with time, and the amplitude and phase of an annual component. The radiosonde data, which are uniformly sampled in time, show an increase in the IPWV of 0.03 mm/yr with a statistical standard deviation of 0.01 mm. The microwave radiometer data, which are not at all uniformly sampled in time, show -0.02¿0.01 mm/yr. We show that the disagreement is caused by the different sampling of the data for the two instruments. When the two data sets are reduced to include only data that are sampled simultaneously, we find an agreement between all estimated model parameters, given their statistical uncertainties. This suggests that if the microwave radiometer had also been operating continuously over the 15-year period, its data would have implied a linear trend similar to the result obtained from the radiosonde data. The general quality of the data, in terms of the short time scatter, has been improved over the time period. The root mean square (RMS) difference between the IPWV measured by the radiometer and by the radiosondes was 2.1 mm during the first 5 years and was reduced to 1.6 mm during the last 4 years. These values include the real difference in the IPWV between the two sites. The bias, radiometer-radiosonde, was 0.1 mm for the whole data set and varied between -0.2 and 0.9 mm for smaller data sets of a few years. ¿ 1998 American Geophysical Union