We measured concentrations and stable carbon isotope compositions of carbon dioxide in the atmospheres of three caves in central Texas and one cave in southern Arizona in order to identify CO2 sources and sinks. The vegetation above the caves studied is either savannah (two caves, above which vegetation has been minimally disturbed) or discrete patches of grassland and woodland (two caves, above which vegetation has been highly disturbed). We tested two hypotheses concerning CO2 in the cave atmospheres: (1) cave ventilation by tropospheric air is the primary sink for CO2 and (2) CO2 is primarily derived from the deepest rooting plants growing above the caves. Within caves, we monitored CO2 at individual locations on monthly and daily time-scales and measured CO2 along transects with increasing distance from the cave entrances. We also measured CO2 in the pore spaces of soils under grasses and trees above each of the caves. We calculated δ13C values of respired CO2 (δ13Cr) for all gas samples using measured δ13C values and CO2 concentrations. We then identified the sources of cave CO2 by comparing cave-air and soil CO2δ13Cr values. At all locations in each Texas cave, CO2 concentrations were highest (lowest) and δ13C values were lowest (highest) during the summer (winter). Cave-air CO2 concentrations consistently increased and δ13C values consistently decreased with distance from the cave entrances. Similar but smaller magnitude seasonal variations in CO2 concentrations occurred in the Arizona cave and no seasonal or spatial variation in the δ13C of cave-air CO2 was observed. The mean δ13Cr values of CO2 in soils under grass were 3.5–4.5‰ higher than the δ13Cr values of CO2 in soils under trees. In the caves under savannah, mean δ13Cr values of cave-air CO2 (−24‰ in both caves) were within 1‰ of the mean δ13Cr values of CO2 in soils under trees. In caves covered by large, contiguous areas of grassland, the δ13Cr values of cave-air CO2 were similar to grassland soil values during the summer and were intermediate between grassland and woodland soil values during the winter. The observed spatial and temporal variations in cave-air CO2 are consistent with density-driven ventilation controlled by seasonal surface temperature changes as the primary sink for CO2 in the Texas caves. The consistent agreement between soil and cave δ13Cr values indicate that the same mixing and diffusion equations that are used to calculate δ13Cr values of soil CO2 also apply to cave-air CO2. Our results suggest that the majority of CO2 advects or diffuses into these caves from soils as a gas rather than being transported in aqueous solution. Measured δ13Cr values and numerical production-diffusion modeling supports our hypothesis that the majority of gaseous CO2 in these caves is derived from deeply rooted vegetation. The carbon isotope composition of groundwater and speleothem calcite used for paleoclimate records are therefore likely biased toward deeply rooted plants, even if sparsely present.