Atmospheric CO₂ concentrations have been obtained from the Atmospheric Infrared Sounder (AIRS) radiance data within the European Centre for Medium-Range Weather Forecasts (ECMWF) data assimilation system. In a first explorative configuration, a subset of channels from the AIRS instrument has been assimilated providing estimates of tropospheric column-averaged CO₂ mixing ratios representative of a layer between the tropopause and about 700 hPa at observation locations only. Results show considerable geographical and temporal variability with values ranging between 370 and 382 ppmv. The 5-day mean estimated random error is about 1%, which is confirmed by comparisons with flask observations on board flights of Japanese airliners in the west-Pacific region. This study demonstrates the feasibility of global CO₂ estimation using high spectral resolution infrared satellite data in a numerical weather prediction data assimilation system. Currently, the system is being improved to treat CO₂ as a full three-dimensional atmospheric variable included in the forecast model. This allows more flexibility in the constraints on the CO₂ estimation as well as the possibility of assimilating other data sources (e.g., near-infrared satellite data and flasks). The CO₂ fields provided by the data assimilation system have great potential to assist the surface flask network in constraining current top-down carbon flux estimates.