Accurate forecasts of very low temperatures in stable boundary layer (SBL) conditions in arctic areas remain a challenge in numerical weather prediction. These forecasts are of great importance both for local communities in exposed areas and in a climate perspective for accurate modeling of changes in permafrost. In this thesis, the Weather Research and Forecasting (WRF) mesoscale meteorological model is evaluated in simulating near-surface temperature in SBL conditions on the Finnmarksvidda plateau in the continental northern Norway. Five temperature inversion episodes are studied in the winters 2015-2017 and in fall 2017 in an area covering the town of Karasjok and the mountain Iškoras. Both locations are equipped with weather stations from the Norwegian Meteorological Institute (MET) measuring quantities such as two-meter temperature (T2m), ten-meter (10-m) wind, shortwave and longwave radiation. The area also has nine stations located at different altitudes between Karasjok and Iškoras measuring T2m. WRF proves to be a useful tool for predicting T2m in SBL conditions, having relatively good correlation with observations. However, for all the studied episodes, WRF overestimates the T2m for the lowest altitude stations during the inversions, while at the stations of highest elevation there is generally a slight underestimation. The cause of the positive temperature biases at the lowest altitudes is found to likely be a result of overestimated 10-m wind speeds and inaccurate longwave radiation fluxes. At the highest altitudes, inaccurate station altitudes from WRF as well as too low 10-m wind speeds limiting advection seem to be the main causes of WRF's deviations in T2m. Downwelling longwave radiation measurements from the MET stations are used as an indication of cloud cover and compared to WRF results, as an accurate representation of cloud cover is crucial for modeling T2m in SBLs during winter. Although often accurately modeled, the cloud cover is found to be simulated poorly during certain episodes, which in turn prevents very stable conditions from developing. It is found that a high horizontal resolution improves WRF's performance as the topography and station altitudes become more accurate. A high vertical resolution is found to be beneficial in most cases, as it improves the vertical temperature gradient. No major differences were found based on the chosen boundary layer (BL) scheme in WRF. The parameterizations of surface fluxes in all the BL schemes and their related surface-layer schemes perform poorly in very stable conditions with non-stationary turbulence, particularly at the lowest altitudes.