The Northern Hemisphere has large areas that are forested and seasonally snow covered. Compared with open areas, forest canopies strongly influence interactions between the atmosphere and snow on the ground by sheltering the snow from wind and solar radiation and by intercepting falling snow, and these influences have important consequences for the meteorology, hydrology and ecology of forests. Many of the land surface models used in meteorological and hydrological forecasting now include representations of canopy snow processes, but these have not been widely tested in comparison with observations. Phase 2 of the Snow Model Intercomparison Project (SnowMIP2) was therefore designed as an intercomparison of surface mass and energy balance simulations for snow in forested areas. Model forcing and calibration data for sites with paired forested and open plots were supplied to modelling groups. Participants in 11 countries contributed outputs from 33 models, and results are published here for sites in Canada, the USA and Switzerland. On average, the models perform fairly well in simulating snow accumulation and ablation, although there is a wide inter-model spread and a tendency to underestimate differences in snow mass between open and forested areas. Most models capture the large differences in surface albedos and temperatures between forest canopies and open snow well. There is, however, a strong tendency for models to underestimate soil temperatures under snow, particularly for forest sites, and this would have large consequences for simulations of runoff and biological processes in the soil.