Due to the fact that the required input data are not always completely available and model structures are only a crude description of the underlying natural processes, model parameters need to be calibrated. Calibrated model parameters only reflect a small domain of the natural processes well. This imposes an obstacle on the accuracy of modelling a wide range of flood events, which, in turn is crucial for flood forecasting systems. Together with the rigid model structures of currently available rainfall-runoff models this presents a serious constraint to portraying the highly non-linear transformation of precipitation into runoff. Different model concepts (interflow, direct runoff), or rather the represented processes, such as infiltration, soil water movement etc. are more or less dominating different sections of the runoff spectrum. Most models do not account for such transient characteristics inherent to the hydrograph. In this paper we try to show a way out of the dilemma of limited model parameter validity. Exemplarily, we investigate on the model performance of WaSiM-ETH, focusing on the parameterisation strategy in the context of flood forecasting. In order to compensate for the non-transient parameters of the WaSiM model we propose a process based parameterisation strategy. This starts from a detailed analysis of the considered catchments rainfall-runoff characteristics. Based on a classification of events, WaSiM-ETH is calibrated and validated to describe all the event classes separately. These specific WaSiM-ETH event class models are then merged to improve the model performance in predicting peak flows. This improved catchment modelling can be used to train an artificial intelligence based black box forecasting tool as described in [Schmitz, G.H., Cullmann, J., Görner, W., Lennartz, F., Dröge, W., 2005. PAI-OFF: Eine neue Strategie zur Hochwasservorhersage in schnellreagierenden Einzugsgebieten. Hydrologie und Wasserbewirtschaftung 49, 226–234; Cullmann, J., Schmitz, G.H., Görner, W., 2006. A new strategy for online flood forecasting in mountainous catchments. in: IAHS Red Book, vol. 303]. Merging of the singular parameter class models is done with the help of a sigmoidal weighting procedure. The new approach thus integrates all available information from the specially calibrated WaSiM-ETH class models, accounting for the different processes and dynamics governing the various event classes. For example it portrays the flood formation process with parameters accounting for the characteristics of the event class models. Implications arising from this study are demonstrated for a catchment in the Erzgebirge (Ore-mountains) in East Germany (1700 km). The computational efficiency, together with the convincing agreement between the predicted and observed flood peaks underlines the potential of the new parameterisation strategy in the context of operational real time forecasting.