Fire regulates the structure and function of savanna ecosystems, yet we lack understanding of how cyclic fire affects savanna carbon dynamics. Furthermore, it is largely unknown how predicted changes in climate may impact the interaction between fire and carbon cycling in these ecosystems. This study utilizes a novel combination of prescribed fire, eddy covariance (EC) and statistical techniques to investigate carbon dynamics in frequently burned longleaf pine savannas along a gradient of soil moisture availability (mesic, intermediate and xeric). This research approach allowed us to investigate the complex interactions between carbon exchange and cyclic fire along the ecological amplitude of longleaf pine. Over three years of EC measurement of net ecosystem exchange (NEE) show that the mesic site was a net carbon sink (NEE = -2.48 tonnes C ha(-1)), while intermediate and xeric sites were net carbon sources (NEE = 1.57 and 1.46 tonnes C ha(-1), respectively), but when carbon losses due to fuel consumption were taken into account, all three sites were carbon sources (10.78, 7.95 and 9.69 tonnes C ha(-1) at the mesic, intermediate and xeric sites, respectively). Nonetheless, rates of NEE returned to pre-fire levels 1-2 months following fire. Consumption of leaf area by prescribed fire was associated with reduction in NEE post-fire, and the system quickly recovered its carbon uptake capacity 30-60 days post fire. While losses due to fire affected carbon balances on short time scales (instantaneous to a few months), drought conditions over the final two years of the study were a more important driver of net carbon loss on yearly to multi-year time scales. However, longer-term observations over greater environmental variability and additional fire cycles would help to more precisely examine interactions between fire and climate and make future predictions about carbon dynamics in these systems.