Ponderosa pine (Pinus ponderosa) forests of the southwestern United States are a mosaic of stands where undisturbed forests are carbon sinks, and stands recovering from wildfires may be sources of carbon to the atmosphere for decades after the fire. However, the relative magnitude of these sinks and sources has never been directly measured in this region, limiting our understanding of the role of fire in regional and US carbon budgets. We used the eddy covariance technique to measure the CO2 exchange of two forest sites, one burned by fire in 1996, and an unburned forest. The fire was a high-intensity stand-replacing burn that killed all trees. Ten years after the fire, the burned site was still a source of CO2 to the atmosphere [109±6 (SEM) g C m−2 yr−1], whereas the unburned site was a sink (−164±23 g C m−2 yr−1). The fire reduced total carbon storage and shifted ecosystem carbon allocation from the forest floor and living biomass to necromass. Annual ecosystem respiration was lower at the burned site (480±5 g C m−2 yr−1) than at the unburned site (710±54 g C m−2 yr−1), but the difference in gross primary production was even larger (372±13 g C m−2 yr−1 at the burned site and 858±37 g C m−2 yr−1at the unburned site). Water availability controlled carbon flux in the warm season at both sites, and the burned site was a source of carbon in all months, even during the summer, when wet and warm conditions favored respiration more than photosynthesis. Our study shows that carbon losses following stand-replacing fires in ponderosa pine forests can persist for decades due to slow recovery of the gross primary production. Because fire exclusion is becoming increasingly difficult in dry western forests, a large US forest carbon sink could shift to a decadal-scale carbon source.