The deterministic Riverstrahler model that simulates nutrient transfer from land-based sources to the sea through the drainage network has been successfully applied to a number of large watersheds ranging from 20,000 to 800,000 km2. In order to assess the feasibility of upscaling such a model in view of its worldwide application based on global databases of hydrology, climate, population, land use, and lithology, a sensitivity analysis has been carried out. The response of the model to differing scales of input data, both high-resolution local data and low-resolution global data, is examined, applying the model to the well-documented Seine River basin. The results show that the evaluation of the overall annual fluxes of nutrients delivered at the outlet of a basin can be correctly predicted even from low resolution input data, provided that (1) the total discharge and seasonal hydrological regime are correctly simulated; (2) the drainage network description does not neglect tributaries that drain more than approximately 4000 km2 watershed area; and (3) the total point sources of nutrients as well as their distribution between the different stream orders are estimated (phosphorus being the most sensitive among the considered nutrients to point sources distribution). Diffuse sources of nutrients can be assessed from low-resolution land use and lithological data. As a result of this study, the capacity of statistical/empirical formulas to assess annual fluxes of nutrient delivery appears comparable to that of deterministic drainage network models, but the latter are particularly suited to predict seasonal variations of riverine nutrient delivery and its elemental ratios.