The abundance of C4 plants on grasslands is controlled by the interplay among atmospheric, climatic, and pedologic factors.  The percentage of C4-to-C3 plants on a grassland decreases with growing season temperature.  Yet experiments suggest that the temperature at which C4 plants might out-compete C3 plants will decrease when atmospheric pCO₂ is lower.  The complexity of these controls has led to considerable debate about the response of grasslands to Quaternary climate change.  Under glacial conditions, was there less C4 biomass (because of lower temperatures) or more C4 biomass (because of lower pCO₂)?  We use isotopic data and coupled climate-vegetation models to assess the balance between C3 and C4 vegetation in Texas during this period.  The carbon isotope composition of fossil bison, mammoth, and horse tooth enamel is a proxy for C3 versus C4 plant consumption, and indicates that C4 plant biomass remained above 55% through most of Texas from prior to the last glacial maximum into the Holocene.  These data also show that horses did not feed exclusively on herbaceous plants, consequently isotopic data from horses are not reliable indicators of the C4-to-C3 balance in grassland biomes.  Estimates of C4 percentages from coupled climate-vegetation models illuminate the relative roles of climate and atmospheric CO₂ concentrations in shaping the regional C4 signal.  C4 percentages estimated using observed modern climate-vegetation relationships and late Quaternary climate variables (simulated by a global climate model) are much lower than those indicated by carbon isotope values from fossils.  When the effect of atmospheric pCO₂ on the competitive balance between C3 and C4 plants is included in the numerical experiment, however, estimated C4 percentages show better agreement with isotopic estimates from late Quaternary mammals and soils.  Our results suggest that low atmospheric CO₂ levels played a role in the observed persistence of C4 plants throughout the late Quaternary.