Fire is an important pathway for carbon (C) loss from boreal forest ecosystems and has a strong effect on ecosystem C balance. Fires can range widely in severity, defined as the amount of vegetation and forest floor consumed by fire, depending on local fuel and climatic conditions. Here we explore a novel method for estimating fire severity and loss of C from fire using the atmosphere to integrate ecosystem heterogeneity at the watershed scale. We measured the δ13C and Δ14C isotopic values of CO2 emitted from an experimental forest fire at the Caribou-Poker Creek Research Watershed (CPCRW), near Fairbanks, Alaska. We used inverse modeling combined with dual isotope measurements of C contained in aboveground black spruce biomass and soil organic horizons to estimate the amount of C released by this fire. The experimental burn was a medium to severe intensity fire that released, on average, about 2.5 kg Cm-2, more than half of the C contained in vegetation and soil organic horizon pools. For vegetation, the model predicted that approximately 70--75% of pools such as needles, fine branches, and bark were consumed by fire, whereas only 20--30% of pools such as coarse branches and cones were consumed. The fire was predicted to have almost completely consumed surface soil organic horizons and burned about half of the deepest humic horizon. The ability to estimate the amount of biomass combusted and C emission from fires at the watershed scale provides an extensive approach that can complement more limited intensive ground-based measurements.