The numerical weather prediction forecast skill of heavy precipitation events in the Mediterranean regions is currently limited, partly because of the paucity of water vapor observations assimilated today. An attempt to fill this observational gap is provided by Global Positioning System (GPS) ground station data over Europe that are now routinely processed into observations of Zenith Total Delay (ZTD), which is closely related to the tropospheric water vapor content. We evaluate here the impact of assimilating the GPS ZTD on the high-resolution (2.4-km) nonhydrostatic prediction of rainfall for the heavy precipitation event of 5–9 September 2005 over Southern France. First, we assimilate the GPS ZTD observations in the three-dimensional variational (3DVAR) data assimilation system of the 9.5-km horizontal resolution ALADIN/France hydrostatic model with parameterized convection. This one-month-long assimilation experiment includes the heavy rainfall period. Prior to the assimilation, a GPS ZTD observation preprocessing is carried out for quality control and bias correction. We find that the GPS ZTD observations impact mainly the representation of the humidity in the low to middle troposphere. We then conduct forecast trials with the Meso-NH model, which explicitly resolves the deep convection, using the analyses of the 3DVAR ALADIN/France assimilation experiments as initial and boundary conditions. Our results indicate a benefit of GPS ZTD data assimilation for improving the Meso-NH precipitation forecasts of the heavy rainfall event.