The surface energy balance at the Svalbard Archipelago has been simulated at high resolution with the Weather Research and Forecasting Model and compared with measurements of the individual energy fluxes from a tundra site near Ny-Ålesund (located north of Norway), as well as other near-surface measurements across the region. For surface air temperature, a good agreement between model and observations was found at all locations. High correlations were also found for daily averaged surface energy fluxes within the different seasons at the main site. The four radiation components showed correlations above 0.5 in all seasons (mostly above 0.9), whereas correlations between 0.3 and 0.8 were found for the sensible and latent heat fluxes. Underestimation of cloud cover and cloud optical thickness led to seasonal biases in incoming shortwave and longwave radiation of up to 30%. During summer, this was mainly a result of distinct days on which the model erroneously simulated cloud-free conditions, whereas the incoming radiation biases appeared to be more related to underestimation of cloud optical thickness during winter. The model overestimated both sensible and latent heat fluxes in most seasons. The model also initially overestimated the average Bowen ratio during summer by a factor of 6, but this bias was greatly reduced with two physically based model modifications that are related to frozen-ground hydrology. The seasonally averaged ground/snow heat flux was mostly in agreement with observations but showed too little short-time variability in the presence of thick snow. Overall, the model reproduced average temperatures well but overestimated diurnal cycles and showed considerable biases in the individual energy fluxes on seasonal and shorter time scales.