Abstract The circulation in fjords with characteristic branched structure and mountainous surroundings are strongly influenced by the local wind pattern. Sufficiently high atmospheric resolution is needed in weather prediction models to resolve the impacts of the topography on atmospheric circulation. We evaluate the effect of using high versus low resolution atmospheric forcing in a numerical ocean model of Porsangerfjorden in northern Norway. For the high resolution case we applied a weather prediction model (WRF-1km) to downscale the atmospheric flow and in the low resolution case we extracted forcing from the global reanalysis ERA-interim. The results show that the reanalysis provides an unrealistic wind pattern within the fjord and a dynamical downscaling is necessary to capture the observed spatial variability in atmospheric circulation. Coarse wind forcing erroneously gives too frequent wind across the fjord, which affects the computed surface circulation and under-predicts eddy activity. A realistic representation of eddy activity is crucial when modeling transport of planktonic organisms. The cod stock in Porsangerfjorden has declined dramatically since the 1970s. We have modeled particle transport within the fjord, identifying retention areas and drift patterns. The results are discussed focusing on historically known cod spawning and nursery areas in the fjord.