Hydrological models sometimes (often?) fail to produce satisfactory simulations. The causes of failures can be manifold: model structure, calibration difficulties, spatial variability, catchment boundary definition, data errors or representativeness, determination of initial conditions, etc. Among these, model structure may be one of the most important sources of uncertainty. However, modelling studies often use a single model, so it is difficult to evaluate the part of the total error due to the model structure only. In case of model failure, should the model be systematically blamed for that? A good way to evaluate the role of model structure is to test a large number of models representing a variety of conceptualisation. This gives an indication of the error that may be due to the choice of the model structure. We adopted this approach on a set of more than 1000 French catchment, on which we tested 17 model structures at the daily time step. Results clearly show that some catchments are outliers only for some models while others are outliers for all. A map of the relative model performances was made to try to identify regional trends in model performance.