It is widely accepted that local and national operational early warning systems can play a key role in mitigating flood damage and losses to society while improving risk awareness and flood preparedness. Over the last years, special attention has been paid to efficiently couple meteorological and hydrological warning systems to track uncertainty and achieve longer lead times in hydrological forecasting. Several national and international scientific programs have focused on the pre-operational test and development of ensemble hydrological forecasting. Based on the lumped soil-moisture-accounting type rainfall-runoff model GRP, developed at Cemagref, we have set up a research tool for ensemble forecasting and conducted several studies to evaluate the quality of streamflow forecasts. The model has been driven by available archives of weather ensemble prediction systems from different sources (Météo-France, ECMWF, TIGGE archive). Our approach has sought to combine overall validation under varied geographical and climate conditions (to assess model robustness and generality) and site-specific validation (to locally accept or reject the hydrologic forecast system and contribute to defining its limits of applicability). The general aim is to contribute to methodological developments concerning a wide range of key aspects in hydrological forecasting, including: the links between predictability skill and catchment characteristics, the magnitude and the distribution of forecasting errors, the analysis of nested or neighbouring catchments for prediction in ungauged basins, as well as the reliability of model predictions when forecasting under conditions not previously encountered during the period of setup and calibration of the system. This presentation will cover the aforementioned topics and present examples from studies carried out to evaluate and inter-compare ensemble forecasting systems using a large and varied set of catchments in France. The specific need to prepare our systems to deal with ungauged basins has led us to work on the processing of ensemble meteorological forecasts together with an ensemble of catchment parameter sets, obtained from regional analysis. Last, we will try to show how progress in this field may be achieved not only by looking at "success stories" or "good catchments", but also by investigating those challenging situations where catchments not reacting to improved methodological efforts can be identified.