Water vapour is a key component of the Earth’s climate system, and its distribution and variability are sources of uncertainty in climate models. The use of long-term integrated water vapour (IWV) observations and reanalyses can help in their assessment. This work pioneered the use of reprocessed GPS IWV data for 1995-2010, converted from estimates of Zenith Total Delay. The conversion was assessed, with the goal of producing a high quality long-term IWV data set. Due to uncertainties in the GPS observations and homogeneity concerns, a global comparison with ERA-Interim reanalysis data was made. Although a general good agreement in means, variability and trends was found, issues in both data sets were highlighted. In GPS, gaps and inhomogeneities in the time series were evidenced, as well as representativeness differences in coastal areas and regions of complex topography. In ERA-Interim, too strong trends in certain regions were found. ERA-Interim was also compared with other reanalyses (MERRA-2, ERA-20C, 20CR), and differences were found in the IWV trends over Africa, Australia, and Antarctica. Finally, GPS and ERA-Interim IWV were used to assess four configurations of the LMDZ atmospheric general circulation model with two different physics and with or without nudging towards ERA-Interim wind fields. Impact of the model physics on the IWV mean was found, with the new physics being moister at tropical latitudes. Overall, the model free runs in both physics have difficulty reproducing the trends and variability observed in ERA-Interim and GPS. This is improved with the nudging, which confirms the importance of large-scale dynamics on IWV trends and variability.