AbstractSupercooled cloud droplets combined with strong wind, can produce heavyice accretions on unheated structures. It is called ”in-cloud atmospheric icing”,and is a well known problem at high elevations in wintertime. Atmosphericicing on wind turbines is a challenge that has to be considered when erectingwind turbines at hills and ridges at high latitudes, for example in Norway.Ice accretion on the turbine blades can reduce the production significantlyand large amounts can stop the turbine entirely. The need of a method topredict atmospheric icing events is increasing since there is a growing interestfor building wind turbines along the windy coastline of Norway.In this study we have tested the ability of a mesoscale numerical weather pre-diction model (Weather Research and Forecasting (WRF) modeling system)to predict in-cloud atmospheric icing events. The simulations were executedwith a fine spatial resolution for a selected area, and with use of a detai-led second-moment parameterization-scheme for the microphysical processes.Two mountains have been used as test sites, Ylläs in Finland and Gamlemsve-ten in Norway, where measurements of icing are available for selected cases.The overall results showed a fairly good agreement between the measure-ments and the simulations for most of the icing events. The experiment atYlläs, where accurate measurements of supercooled cloud water were direct-ly compared to the modeled cloud water content, gave the best results. Theratio between modeled and measured values was about 1.3 for all the casesin the finest grid, and about 0.8 in all the cases when the model’s spatialresolution was decreased by a factor 4.The results from the experiment at Gamlemsveten are a bit more intricateto analyze because the modeled ice loads, which is compared to the measu-rements, is calculated from temperature and wind speed in addition to cloudwater. There are also several uncertainties regarding the comparison betweenthe calculated and the observed ice loads. The agreement between the mo-deled and the observed ice loads seems to be best in weather situations withlow stratus clouds containing mostly liquid cloud water. The model seems tounderestimate the icing rate in a period with convective clouds in cold airmasses, when cloud ice is mixed into the cloud.