Model predictions of biogeochemical fluxes on the landscape scale are highly uncertain, both with respect to stochastic (parameter) and structural uncertainty. The idea of our ensemble modelling approach is to reduce the predictive uncertainty by covering part of the parameter and model structural uncertainty. In this study 4 different models (LASCAM, a modified INCA model, SWAT and HBV-N-D) designed to simulate hydrological fluxes as well as mobilization and transport of one or several nitrogen species are applied over the meso-scaled River Fyris catchment in Mid-Eastern Sweden. Hydrological calibration against 5 years of recorded discharge at two stations gives highly variable results from Nash-Sutcliffe Efficiency (NSE) values above 0.80 to values around 0.50. SWAT and HBV-N-D gives alternatively the best simulation result at each station respectively. Alteration of nitrogen parameters following Monte-Carlo or Latin-Hypercube stratified sampling schemes is realized in order to cover the parameter uncertainty of predictions for 3 nitrogen species: nitrate (NO3), ammonium (NH4) and total nitrogen (Tot-N) in terms of exported loads. For each model and each nitrogen species, predictions are ranked in two different ways regarding the performance indicated by two different objective functions: the coefficient of determination R2 and the Nash-Sutcliffe Efficiency (NSE). Model ensembles were compiled in various ways. A total of 396 Single Model Ensembles (SME) are generated using an increasing number of model members. Finally, 78 Multi-Model Ensembles (MME) are combined by using the best SME for each model, nitrogen species and station. The evolution of the two aforementioned objective functions is used as performance descriptor of the ensemble procedure. In each studied case, there is always at least one compilation scheme which outperforms any of its members. The best SME are multiple-linear regression models with R2 selected members, increasing the best NSE values from negativity up to very high ones (0.83). The uncertainty bounds of the SME are almost always smaller than the one introduced by the whole set of selected single model runs still including most of measurements and even more (half of the cases) than the bounds of the selected single runs set. In the same way, there is always at least one MME combination scheme which outperforms all the SME, but the increase in model performance is pronounced than the difference between single model runs and SME. The best MME are the ones with the most members and both R2 and NSE values are reaching 0.89 in the best case. Uncertainty areas described by MME are alternatively increased or reduced compared to the bounds delineated by their members. No general trend is deduced for the studied cases.@en