Improved representations of snow interception by coniferous forest canopies and sublimation of intercepted snow are implemented in a land surface model. Driven with meteorological observations from forested sites in Canada, the United States, and Sweden, the modified model is found to give reduced sublimation, better simulations of snow loads on and below canopies, and improved predictions of snowmelt runoff. When coupled to an atmospheric model in a GCM, however, drying and warming of the air because of the reduced sublimation provides a feedback that limits the impact of the new canopy snow model on the predicted sublimation. There is little impact on the average annual snowmelt runoff in the GCM, but runoff is delayed and peak runoff increased by the introduction of the canopy snow model.