This study assessed the impact of grazing, with and without cattle, on carbon pools in an Australian semi-arid woodland using field measurements and simulations of the Full Carbon Accounting Model (FullCAM). Field measurements included living and dead trees, shrubs and grasses, litter and coarse woody debris (aboveground) and biocrust and soil to 10 cm depth (belowground). Root biomass was estimated as a shoot-to-root ratio. Total carbon (above + below ground) in cattle-excluded grazed sites (75.0 +/- 0.08 t ha-1) was similar to cattle-included grazed sites (72.4 +/- 0.08 t ha-1), with approximately 16-21 % of the total measured carbon stored in the soil. However, cattle grazing reduced biocrust carbon from 0.24 t ha-1 to 0.06 t ha-1, emphasizing the vulnerability of biocrusts to hard-hooved trampling by large and heavy cattle. The field measurements were used to calibrate model parameters within FullCAM so that simulated carbon pools aligned with field-measured carbon stocks. Following the calibration step, grazing scenarios, with and without cattle, were simulated from 1900 to 2300, including periodic wildfire occurrences. These simulations showed lower live tree carbon stock in cattle-included grazing treatment that led to a gradual decline in total carbon. The total carbon stock under the cattle-excluded grazing scenario remained between 90 and 103 t ha-1 from 1900 to 2100, but decreased to 86-100 t ha-1 following the introduction of cattle-included grazing conditions. Our results highlight the degrading impacts of cattle-included grazing on biocrusts, which in turn undermine soil stability and carbon retention. The application of FullCAM to simulate carbon pools in semi-arid woodland ecosystems over decades and centuries extends the interpretation of results from short-term studies in plausible ways.