Should flood forecasting models be continuous or event-based? This question is still a matter of debate in the hydrological community and it seems that no general answer was provided so far. Does the choice of one or the other strategy significantly impact the quality and robustness of the forecasts? Can it be responsible for the failure of a forecasting model? The objective of this study is to bring some new insights in this debate by overtaking the limits of previous studies that were carried out on a single or a limited number of catchments. This study aims to compare two different types of flood forecasting approaches. The first one runs on a continuous mode (i.e. initial conditions are determined only at the beginning of the time series), while the second one is event based (i.e. initial conditions are determined before each flood event). To test these strategies, we used the lumped hydrological models developed at Cemagref over the last decades. They were applied at the hourly time step on a large set of about 1000 French catchments to get general conclusions. These catchments are various in size (10 to 10000 km²) and cover a wide range of hydro-climatic conditions. A common framework was used to evaluate the two approaches in the same conditions. Results were mapped to try to establish regional trends. From an operational point of view, this should help flood forecasting services to choose which initialization strategy to use in their particular interest area (i.e. continuous or event based) to limit model failures. This should also be matter for thought for the future directions of improvement of flood forecasting models.