Data from two different sensors measuring ice particles were combined to establish an improved data series for ice water content. Together with liquid water measurements this new data set was used to evaluate arctic cloud properties, simulated with a state-of-the-art mesoscale atmospheric model. Dependent on the method used for comparison, the mean cloud fraction was found to lie between 51 and 58% in the observations and 53% in the simulations. The hit rate for the total water content was estimated to be between 65 and 71% and the systematic error to a few percent. On the basis of in-cloud observations only, the model was able to reproduce cloudy conditions in 74% of the data points. On the other hand, the model underestimated the occurrence of the liquid phase by about 80% and slightly overestimated the occurrence of the ice phase. Also, in the temperature range from 255 K to 230 K, where considerable amounts of supercooled water were observed, the model failed to produce the liquid phase. Our results confirm the previous finding that despite high forecast skill with respect to cloudy or cloud-free events, the model underpredicts the occurrence of liquid phase in arctic clouds. This shortcoming will have a large influence on precipitation forecasts, as well as on climate predictions. Despite some improvements in recent years, more research is needed to improve the parameterization of arctic cloud properties in fine-scale weather prediction models. For climate models, which have to employ a much cruder parameterization of the microphysics, we face a number of challenges.