The use of weather ensemble predictions in ensemble flood forecasting is an acknowledged procedure to include the uncertainty of meteorological forecasts in a probabilistic streamflow prediction system. Operational flood forecasters can thus get an overview of the probability of exceeding a critical discharge or water level, and decide on whether a flood warning should be issued or not. This process offers several challenges to forecasters: 1) how to define critical thresholds along all the rivers under survey? 2) How to link locally defined thresholds to simulated discharges, which result from models with specific spatial and temporal resolutions? 3) How to define the number of ensemble forecasts predicting the exceedance of critical thresholds necessary to launch a warning? This study focuses on the third challenge. We investigate the optimal number of ensemble members exceeding a critical discharge in order to issue a flood warning. The optimal ensemble threshold is the one that minimizes the number of false alarms and misses, while it optimizes the number of flood events correctly forecasted. Furthermore, in our study, an optimal ensemble threshold also maximizes flood preparedness: the gain in lead-time compared to a deterministic forecast. Data used to evaluate critical thresholds for ensemble flood forecasting come from a selection of 208 catchments in France, which covers a wide range of the hydroclimatic conditions (including catchment size) encountered in the country. The GRPE hydrological forecasting model, a lumped soil-moisture-accounting type rainfall-runoff model, is used. The model is driven by the 10-day ECMWF deterministic and ensemble (51 members) precipitation forecasts for a period of 18 months. A trade-off between the number of hits, misses, false alarms (Critical Success Index) and the gain in lead time is sought to find the optimal number of ensemble members exceeding the critical discharge. In this study the focus lies on the start of a flood event, since the first day with a threshold exceedance is the most important and the most difficult to forecast. The results show us that there is no overall ensemble threshold for the streamflow predictions based on the ECWMF forecast which results in higher CSI and a gain in lead-time for the exceedance of the Q99 streamflow threshold (99th percentile over the same period of 18 months) compared to the deterministic forecast. The optimal ensemble streamflow predictions for the lower streamflow thresholds results also in a negative preparedness score (i.e. loss in lead-time). The same analysis is conducted for a sub selection consisting of 29 large catchments all over France. The results of this analysis show us that in this case there is an ensemble threshold for the Q99 streamflow thresholds which results in a higher CSI score and a gain in preparedness (i.e. gain in lead-time) compared to the deterministic forecast. Both scores, the CSI and preparedness, could be maximized when a catchment specific ensemble threshold is applied. These optimal ensemble thresholds are further explored in order to search for correlations with catchment characteristics, forecast lead-time and discharge thresholds.