Flash floods frequently hit Southern France and cause heavy damages and fatalities. To better protect persons and goods, official flood forecasting services in France need accurate information and efficient models to optimize their decision and policy. Since heavy rainfalls that cause such floods are very heterogeneous , it becomes a serious challenge for forecasters. Such phenomena are typically nonlinear and more complex than classical floods events. That problem leads to consider complementary alternatives to enhance the management of such situations. For decades, artificial neural networks have been very efficient to model nonlinear phenomena, particularly rainfall-discharge relations in various types of basins. They are applied in this study with two main goals: first modelling flash floods on the Gardon de Mialet basin; second, extract internal information from the model by using the Knowledge eXtraction method to provide new ways to improve models. The first analysis shows that the kind of nonlinear predictor influences strongly the representation of information: e.g. the main in-fluent variable (rainfall) is more important in the recurrent and static models than in the feed-forward one. For understanding flash floods genesis, recurrent and static models appear thus as better candidates, even if their results are better.