The role of surface freshwater flux boundary conditions in prognostic Arctic Ocean/sea-ice modelling is analysed, using a regional numerical model. Three different applications of freshwater flux formulations are evaluated. The standard formulation, which serves as a benchmark, takes surface volume fluxes due to precipitation, evaporation and river runoff into account. The total freshwater input to the Arctic Ocean by runoff, precipitation and Bering Strait inflow is approximately 6800 km3 yr-1 in the model setup. The implementation of an Arctic river water tracer in the standard run enables the calculation of an average mean residence time of 14-15 years for river water in the Arctic halocline. The second formulation for surface freshwater fluxes neglects the volume input, which corresponds to applying 'virtual salinity fluxes'. This simplification leads to a rapid salinity build-up in the upper layers of the Arctic Ocean and causes a substantial reduction of freshwater export through Fram Strait. The third formulation uses a constant reference salinity of 35 psu in the definition of the virtual salinity flux boundary condition. This approach results in hydrographic fields which are very similar to those from the standard run. Errors in circulation and freshwater transport are small and, for most applications, tolerable. Our results suggest that virtual salinity fluxes with fixed reference salinity are a reasonable approximation for Arctic Ocean models with horizontal resolution of order 100 km. © 2005 Elsevier Ltd. All rights reserved.