The spatial variability of near-surface variables and turbulent surface fluxes was investigated in three Arctic fjords in Svalbard applying the Weather Research and Forecasting (WRF) mesoscale model. Ten real cases from winter and spring 2008, representing the most common large-scale flow directions, were simulated at 9, 3 and 1 km resolutions for 36 h each. Validation against tower observations and radiosoundings showed fairly good agreement, although a systematic warm and moist bias and slightly overestimated wind speeds were found close to the surface. The spatial variability within a fjord was large and it often reached levels comparable to the temporal variability. The spatial variability of the surface fluxes of sensible and latent heat was mostly controlled by the air and sea surface temperatures instead of wind speed. The same cases were also simulated without any topography over Svalbard. The topography increased the spatial variability but the influence on the mean values was not systematic, except that a clear warming effect was seen in all the fjords studied. The role of surface type increased with increasing air–sea temperature difference and was dominating over topographic effects for the air temperature, specific humidity and turbulent heat fluxes.