Abstract. Measurements of snowmelt and turbulent heat fluxes were made during the snowmelt periods of two years at two neighbouring tundra sites in the Yukon, one in a sheltered location with tall shrubs exposed above deep snow and the other in an exposed location with dwarf shrubs covered by shallow snow. The snow was about twice as deep in the valley as on the plateau at the end of each winter and melted out about 10 days later. The site with buried vegetation showed a transition from air-to-surface heat transfers to surface-to-air heat transfers as bare ground became exposed during snowmelt, but there were daytime transfers of heat from the surface to the air at the site with exposed vegetation even while snow remained on the ground. A model calculating separate energy balances for snow and exposed vegetation, driven with meteorological data from the sites, is found to be able to reproduce these behaviours. Averaged over 30-day periods the model gives about 8 Wm−2 more sensible heat flux to the atmosphere for the valley site than for the plateau site. Sensitivity of simulated fluxes to model parameters describing vegetation cover and density is investigated.