Flash flooding events are typically triggered by extreme rainfall and efficient runoff production and are thus difficult to monitor and predict at the spatial and temporal scales of interest due to large meteorological and hydrologic uncertainties. To provide flash flood warnings for small ungauged basins, Météo-France and Irstea (formerly Cemagref) have developed a discharge-threshold flood warning system called AIGA, which combines the operational radar-gauge rainfall grids with a simplified distributed hydrologic model run every 15 minutes at a 1 -km² resolution. Operational since 2005 in the South of France, AIGA produces in real-time peak discharge estimates along the river network, which are compared to regionalized flood frequency estimates. Warnings are then provided to the French regional and national flood forecasting centres based on the range of the AIGA-estimated return period of ongoing events. The rainfall-runoff model is currently being enhanced to apply AIGA to the entire French territory by 2016. To further extend the forecast lead time, deterministic rainfall forecasts from Météo-France’s Application of Research to Operations at Mesoscale (AROME) convection-permitting model are ingested as time-lagged input ensembles in the hydrologic model. Operational since December 2008, AROME produces every six hours 30-h forecasts at a 2.5-km resolution to resolve deep convection processes. Deterministic AROME forecasts for the Meuse and Moselle river basins were provided for 20 significant rain events between January 2009 and December 2012. The time-lagged approach is a practical choice of accounting for the atmospheric forecast uncertainty when no extensive forecast archive is available for statistical modelling. It enabled us to produce AROME-based streamflow ensembles up to a 12-h forecast horizon and demonstrate the potential improvements in terms of flash flood event detection and effective lead-time. Warnings were also derived by comparing AROME-timelagged precipitation forecasts with Flash Flood Guidance (FFG) values. Results for the 39 studied basins showed that the FFG approach, which does not account for the rainfall spatial and temporal distribution within the basin, did not degrade the quality of flash flood warnings.