We present temperature inversion characteristics during fog and nonfog conditions at three east Greenland coastal weather stations during Arctic melt seasons 1980–2016. For this purpose, we developed a novel automated method to extract fog‐top height (FTH) from Integrated Global Radiosonde Archive data, which is applicable to any fog thermodynamic profile and includes an improved interpolation of saturation between sounding levels. From the analysis of >22,000 melt‐season soundings we conclude that inversions occur 85–95% of the time, are predominantly elevated, and have median depths >200 m. Fog at high‐Arctic locations often penetrates the inversion layer, especially in the late melt season, and is commonly several hundred meters thick. At low‐Arctic locations fog is thinner and generally restricted to the mixed layer. Inversions during fog are deeper and stronger compared to nonfog conditions. This effect is more pronounced at higher latitudes, which we attribute to distinct local boundary layer conditions and large‐scale processes. The Integrated Global Radiosonde Archive‐extracted FTHs have a cumulative error of 56 m and are in reasonable agreement with retrievals from Cloud‐Aerosol Lidar and Infrared Pathfinder Satellite Observation satellite cloud top data. The novel FTH extraction method can be applied to any polar sounding with >5 significant levels below 700 hPa and can be extended to boundary layer clouds other than fog, which represent the majority of cloud occurrence in the Arctic melt season. This study advances the understanding of interactions between low clouds and temperature inversions and improves retrieval of cloud geometrical thickness from radiosondes: both have important implications for the Arctic surface energy budget.