Résumé : The river–floodplain network plays an important role in the carbon (C) cycle of the Amazon basin, as it transports and processes a significant fraction of the C fixed by terrestrial vegetation, most of which evades as CO 2 from rivers and floodplains back to the atmosphere. There is empirical evidence that exceptionally dry or wet years have an impact on the net C balance in the Amazon. While seasonal and interannual variations in hydrology have a direct impact on the amounts of C transferred through the river–floodplain system, it is not known how far the variation of these fluxes affects the overall Amazon C balance. Here, we introduce a new wetland forcing file for the ORCHILEAK model, which improves the representation of floodplain dynamics and allows us to closely reproduce data-driven estimates of net C exports through the river–floodplain network. Based on this new wetland forcing and two climate forcing datasets, we show that across the Amazon, the percentage of net primary productivity lost to the river–floodplain system is highly variable at the interannual timescale, and wet years fuel aquatic CO 2 evasion. However, at the same time overall net ecosystem productivity (NEP) and C sequestration are highest during wet years, partly due to reduced decomposition rates in water-logged floodplain soils. It is years with the lowest discharge and floodplain inundation, often associated with El Nino events, that have the lowest NEP and the highest total (terrestrial plus aquatic) CO 2 emissions back to atmosphere. Furthermore, we find that aquatic C fluxes display greater variation than terrestrial C fluxes, and that this variation significantly dampens the interannual variability in NEP of the Amazon basin. These results call for a more integrative view of the C fluxes through the vegetation-soil-river-floodplain continuum, which directly places aquatic C fluxes into the overall C budget of the Amazon basin.