Terrestrial ecosystems control carbon dioxide fluxes to and from the atmosphere through photosynthesis and respiration, a balance between net primary productivity and heterotrophic respiration, that determines whether an ecosystem is sequestering carbon or releasing it to the atmosphere. Global and site-specific data sets have demonstrated that climate and climate variability influence biogeochemical processes that determine net ecosystem carbon dioxide exchange (NEE) at multiple timescales. Experimental data necessary to quantify impacts of a single climate variable, such as temperature anomalies, on NEE and carbon sequestration of ecosystems at interannual timescales have been lacking. This derives from an inability of field studies to avoid the confounding effects of natural intra-annual and interannual variability in temperature and precipitation. Here we present results from a four-year study using replicate 12,000-kg intact tallgrass prairie monoliths located in four 184-m(3) enclosed lysimeters. We exposed 6 of 12 monoliths to an anomalously warm year in the second year of the study and continuously quantified rates of ecosystem processes, including NEE. We find that warming decreases NEE in both the extreme year and the following year by inducing drought that suppresses net primary productivity in the extreme year and by stimulating heterotrophic respiration of soil biota in the subsequent year. Our data indicate that two years are required for NEE in the previously warmed experimental ecosystems to recover to levels measured in the control ecosystems. This time lag caused net ecosystem carbon sequestration in previously warmed ecosystems to be decreased threefold over the study period, compared with control ecosystems. Our findings suggest that more frequent anomalously warm years, a possible consequence of increasing anthropogenic carbon dioxide levels, may lead to a sustained decrease in carbon dioxide uptake by terrestrial ecosystems.