Atmospheric water vapour plays a key role in the Arctic radiation budget, hydrological cycle and hence climate, but its measurement with high accuracy remains an important challenge. Total column water vapour (TCWV) datasets derived from ground-based GNSS measurements are used to assess the quality of different existing satellite TCWV datasets, namely from the Moderate Resolution Imaging Spectroradiometer (MODIS), the Atmospheric Infrared Sounder (AIRS) and the SCanning Imaging Absorption spectroMeter for Atmospheric CHartographY (SCIAMACHY). The comparisons between GNSS and satellite data are carried out for three reference Arctic observation sites (Sodankylä, Ny-Ålesund and Thule) where long homogeneous GNSS time series of more than a decade (2001–2014) are available. We select hourly GNSS data that are coincident with overpasses of the different satellites over the three sites and then average them into monthly means that are compared with monthly mean satellite products for different seasons. The agreement between GNSS and satellite time series is generally within 5 % at all sites for most conditions. The weakest correlations are found during summer. Among all the satellite data, AIRS shows the best agreement with GNSS time series, though AIRS TCWV is often slightly too high in drier atmospheres (i.e. high-latitude stations during autumn and winter). SCIAMACHY TCWV data are generally drier than GNSS measurements at all the stations during the summer. This study suggests that these biases are associated with cloud cover, especially at Ny-Ålesund and Thule. The dry biases of MODIS and SCIAMACHY observations are most pronounced at Sodankylä during the snow season (from October to March). Regarding SCIAMACHY, this bias is possibly linked to the fact that the SCIAMACHY TCWV retrieval does not take accurately into account the variations in surface albedo, notably in the presence of snow with a nearby canopy as in Sodankylä. The MODIS bias at Sodankylä is found to be correlated with cloud cover fraction and is also expected to be affected by other atmospheric or surface albedo changes linked for instance to the presence of forests or anthropogenic emissions. Overall, the results point out that a better estimation of seasonally dependent surface albedo and a better consideration of vertically resolved cloud cover are recommended if biases in satellite measurements are to be reduced in the polar regions.