This study assesses the quality of streamflow forecasts issued by the GRP rainfall-runoff model, driven by ensemble weather predictions from the TIGGE database. The GRP is a lumped soil-moisture-accounting type model developed at Cemagref in France for operational flood forecasting. In this study, the model is run at daily time steps and its assimilation procedure makes use of the last observed discharge at the time of the forecast to update the state of the model routing store. Ensemble predictions from 8 meteorological centres from the TIGGE archive are considered. They are available over a 2-year period, from October 2006 to October 2008. Single ensembles from each centre, with a number of members ranging from 15 to 51 ensembles, are considered in the hydrological model, as well as a combined multi-model ensemble that takes into account all members available at each day of the forecast period. Forecast data were spatially averaged over the studied catchments to obtain the areal forecast precipitation at each lead time. Hydrological forecast evaluation is performed at catchment-based spatial scales and for lead-times up to 15 days. Simulations were carried out over 74 catchments in France, with areas ranging from 1000 km² to 44,000 km². Forecasts are compared to observed data and typical skill scores used for forecast verification are computed. Issues related to the quality of the ensemble hydrological forecasts, and the impacts of factors like catchment area and hydrological regime type on the performance of the forecasts are discussed.