Models used to estimate net ecosystem productivity (NEP) during climate change should first be shown to simulate the effects on NEP of interannual variability in current climates. Energy and CO2 fluxes simulated by the ecosystem model ecosys were compared with those measured by eddy covariance over a semi-arid ungrazed grassland near Lethbridge, Alberta to improve confidence in model projections of grassland NEP. Differences between simulated and measured fluxes of CO2 and energy were within the range of uncertainty in measured fluxes during 3 years with declining precipitation (466, 363, and 276 mm in 1998, 1999, and 2000, respectively). Standard differences between modelled versus measured CO2 and LE over 3 years were 1.0 μmol m−2 s−1 and 22 W m−2 which were comparable to standard errors estimated from eddy covariance measurements. However, further research is needed to reconcile differences in modelled vs. measured ecosystem respiration. Annual modelled NEP of the semi-arid ungrazed grassland was +59 g C m−2 in 1998, +5 g C m−2 in 1999, and −33 g C m−2 in 2000. Long-term modelled rate of C accumulation in this grassland under current climate was 26 g C m−2 y−1 which is consistent with C accumulation in semiarid grasslands estimated from measurements and models elsewhere. Under climate change projected by the Canadian Regional Climate Model II from the IS92a emissions scenario, increases in transpiration caused by rising temperatures were fully offset by decreases in transpiration caused by rising CO2, thereby alleviating water deficits and lengthening growing seasons. The consequent rise in grassland net primary productivity was largely offset by a rise in heterotrophic respiration, so that C sequestration by the semi-arid ungrazed grassland rose by less than 2 g C m−2 y−1 under climate change.