Abstract The impacts of grazing animals on the carbon balance of grasslands globally can potentially impact atmospheric carbon dioxide concentrations and soil carbon stocks. In this study, we measured net ecosystem CO2 exchange (FN) for two periods with contrasting environmental variables, in early and late summer. We used a closed dynamic chamber system for continuous measurements of the components of FN (gross canopy photosynthesis, A, and ecosystem respiration, RE, and its components: above-ground plant respiration, RL and soil respiration, RS). We calculated an empirical phytomass index (calculated Pi) from continuous measurements of FN to account for variability in above-ground biomass following grazing events. In addition, we developed a model for the phytomass index from environmental variables (modelled Pi’). Then, using cross validations, we tested the efficacy of calculated Pi and modelled Pi’ in improving environmental models for FN and its components. Grazing events affected A and RL significantly but did not result in changes in RS. Inclusion of the phytomass index (either calculated or modelled) increased the predictive performances of the models for A and RL, but not for RS. For both measurement periods, the ecosystem was a sink of CO2 with mean ± standard error (SE) rates of net uptake of 1.88 ± 0.23 and 1.47 ± 0.69 g C m−2 d−1 for the periods in early summer and in late summer, respectively. Overall, FN was poorly described by the models when the phytomass indices were not included. Estimates of FN using either the calculated or the modelled phytomass index over both periods were not significantly different from the measured values. We showed that the use of a modelled phytomass index can account for the effects of grazing well, and can be used to improve models to predict FN and the carbon balance for grazed grasslands under different climatic conditions and management practices.