Rivers and streams are control points for CO2 emission to the air (fCO2), with emission rates often exceeding internal metabolism (net ecosystem production, NEP). The difference is usually attributed to CO2-supersaturated groundwater inputs from upland soil respiration and rock weathering, but this implies a terrestrial-to-aquatic C transfer greater than estimated by terrestrial mass balance. One explanation is that riparian zones—rich in organic and inorganic C but mostly neglected in terrestrial mass balances—contribute disproportionately to fCO2. To test this hypothesis, we measured fCO2, NEP, and the lateral CO2 contributions from both terrestrial uplands (TER) and riparian wetlands (RIP) for seven reaches in a lowland river network in Florida, USA. NEP contributed about half of fCO2, but the remaining CO2 emission was generally much larger than measured TER. The relative importance of RIP versus TER varied markedly between contrasting hydrogeologic settings; RIP contributed 49% of fCO2 where geologic confinement forced lateral drainage through riparian soils, but only 12% where unconfined karst allowed deeper groundwater flowpaths that bypassed riparian zones. On a land area basis, the narrow riparian corridor yielded far more CO2 than the terrestrial uplands (33.1 vs. 1.4 g-C m−2 yr−1), resulting in river corridors (i.e., stream channel plus adjacent wetlands, NEP + RIP) sourcing 87% of fCO2 to streams. Our findings imply that true terrestrial CO2 subsidies to streams may be smaller than previously estimated by aquatic mass balance and highlight the importance of explicitly integrating riparian zones into the conceptual model for terrestrial-to-aquatic C transfer.