Various types of precipitable water (PW) measurement are compared for different sites around Tucson, Arizona, where arid conditions prevail, and the sensitivity of irradiance to PW variations is largest. The accuracy of some determinations of this quantity is assessed by comparison with routine GPS meteorology data. Large scatter is obtained with all types of empirical functions relating PW to surface temperature and humidity data, but the climate sensitivity of this kind of determination is found lowest when relating PW to the surface specific humidity, rather than the more usual vapor pressure or dew point temperature. The impact on the accuracy of predicted direct normal irradiance (DNI) and global horizontal irradiance (GHI) of various sources of PW data, at either low or high temporal resolution, is assessed using predictions from the REST2 radiative model, in combination with co-located sunphotometric and radiometric data at Tucson during a 7-month period. Results suggest that the accuracy of the predicted DNI and GHI is only lightly sensitive to the uncertainty in the input PW data. In case PW is not measured locally, a convenient source of data is provided by reanalysis, such as from the MERRA model. (C) 2013 Elsevier Ltd. All rights reserved.