Rising air temperatures are believed to be hastening heterotrophic respiration (Rh) in arctic tundra ecosystems, which could lead to substantial losses of soil carbon (C). In order to improve confidence in predicting the likelihood of such loss, the comprehensive ecosystem model ecosys was first tested with carbon dioxide (CO2) fluxes measured over a tundra soil in a growth chamber under various temperatures and soil-water contents (θ). The model was then tested with CO2 and energy fluxes measured over a coastal arctic tundra near Barrow, Alaska, under a range of weather conditions during 1998–1999. A rise in growth chamber temperature from 7 to 15 °C caused large, but commensurate, rises in respiration and CO2 fixation, and so no significant effect on net CO2 exchange was modelled or measured. An increase in growth chamber θ from field capacity to saturation caused substantial reductions in respiration but not in CO2 fixation, and so an increase in net CO2 exchange was modelled and measured. Long daylengths over the coastal tundra at Barrow caused an almost continuous C sink to be modelled and measured during most of July (2–4 g C m−2 d−1), but shortening daylengths and declining air temperatures caused a C source to be modelled and measured by early September (∼1 g C m−2 d−1). At an annual time scale, the coastal tundra was modelled to be a small C sink (4 g C m−2 y−1) during 1998 when average air temperatures were 4 °C above normal, and a larger C sink (16 g C m−2 y−1) during 1999 when air temperatures were close to long-term normals. During 100 years under rising atmospheric CO2 concentration (Ca), air temperature and precipitation driven by the IS92a emissions scenario, modelled Rh rose commensurately with net primary productivity (NPP) under both current and elevated rates of atmospheric nitrogen (N) deposition, so that changes in soil C remained small. However, methane (CH4) emissions were predicted to rise substantially in coastal tundra with IS92a-driven climate change (from ∼20 to ∼40 g C m−2 y−1), causing a substantial increase in the emission of CO2 equivalents. If the rate of temperature increase hypothesized in the IS92a emissions scenario had been raised by 50%, substantial losses of soil C (∼1 kg C m−2) would have been modelled after 100 years, including additional emissions of CH4.