The application of stationary parameters in conceptual hydrological models, even under changing boundary conditions, is a common yet unproven practice. This study investigates the impact of non-stationary model parameters on model performance for different flow indices and time scales. Therefore, a Self-Organizing Map based optimization approach, which links nonstationary model parameters with climate indices, is presented and tested on seven meso-scale catchments in northern Germany. The algorithm automatically groups sub-periods with similar climate characteristics and allocates them to similar model parameter sets. The climate indices used for the classification of sub-periods are based on (a) yearly means and (b) a moving average over the previous 61 days. Classification b supports the estimation of continuous non-stationary parameters. The results show that (i) non-stationary model parameters can improve the performance of hydrological models with an acceptable growth in parameter uncertainty; (ii) some model parameters are highly correlated to some climate indices; (iii) the model performance improves more for monthly means than yearly means; and (iv) in general low to medium flows improve more than high flows. It was further shown how the gained knowledge can be used to identify insufficiencies in the model structure. Figure 12. Yearly maximum runoff (left figure) and mean monthly maximum runoffs of the validation period (right figure) of the Böhme catchment (Br) for observations (Obs), simulations with stationary parameters (St), and simulations with non-stationary parameters (NSt) 3157 NON-STATIONARY HYDROLOGICAL MODEL PARAMETERS BASED ON SOM-B