Based on about 20 years of NOAA/CMDL's atmospheric CO₂ concentration data and a global atmospheric tracer transport model, we estimate interannual variations and spatial patterns of surface CO₂ fluxes in the period 01/1982-12/2000, by using a time-dependent Bayesian inversion technique. To increase the reliability of the estimated temporal features, particular care is exerted towards the selection of data records that are homogeneous in time. Fluxes are estimated on a grid-scale resolution (~8º latitude x 10º longitude), constrained by a-priori spatial correlations, and then integrated over different sets of regions. The transport model is driven by interannually varying re-analyzed meteorological fields. We make consistent use of unsmoothed measurements. In agreement with previous studies, land fluxes are estimated to be the main driver of interannual variations in the global CO₂ fluxes, with the pace predominantly being set by the El Niño/La Niña contrast. An exception is a 2-3 year period of increased sink of atmospheric carbon after Mt.  Pinatubo's volcanic eruption in 1991. The largest differences in fluxes between El Niño and La Niña are found in the tropical land regions, the main share being due to the Amazon basin. The flux variations for the Post-Pinatubo period, the 1997/1998 El Niño, and the 1999 La Niña events are exploited to investigate relations between CO₂ fluxes and climate forcing. A rough comparison points to anomalies in precipitation as a prominent climate factor for short-term variability of tropical land fluxes, both through their role on NPP and through promoting fire in case of droughts. Some large flux anomalies seem to be directly related to large biomass burning events recorded by satellite observation. Global ocean carbon uptake shows a trend similar to the one expected if ocean uptake scales proportional to the anthropogenic atmospheric CO₂ perturbation. In contrast to temporal variations, the longterm spatial flux distribution can be inferred with lesser robustness only. The tentative pattern estimated by the present inversion exhibits a northern hemisphere land sink on the order of 0.4 PgC/yr (for 01/1996-12/1999, non-fossil fuel carbon only) that is mainly confined to North America. Southern hemisphere land regions are carbon neutral, while the tropical land regions are taking up carbon (e.g., at a rate of 0.8 PgC/yr during 01/1996-12/1999). Ocean fluxes show larger uptake in the Northern mid to high latitudes than in the Southern mid latitude regions, in contrast to the estimates by Takahashi et al. (1999) based on in-situ measurements. On a regional basis, results that differ the most from previous estimates are large carbon uptake of 1 to 1.5 PgC/yr by the Southern temperate Pacific ocean region, weak outgassing from the Southern ocean, and a carbon source from eastern Europe.