Abstract. The magnitude of the interannual variability of North Atlantic air-sea CO₂ fluxes remains uncertain. Interannual extremes simulated by atmospheric inverse approaches are typically about ±0.3 Pg C yr⁻¹, whereas those from ocean models are less than ±0.1 Pg C yr⁻¹. Thus variability in the North Atlantic is either about 60% or less than 20% of the global variability of about ±0.5 Pg C yr⁻¹ (as estimated by both approaches). Here we explore spatiotemporal variability within the North Atlantic basin of one ocean model in order to more fully describe potential counteracting trends in different regions that may explain why basin-wide variability is small relative to global-scale variability. Typical atmospheric inverse approaches separate the North Atlantic into at most a few regions and thus cannot properly simulate such counteracting effects. For this study, two simulations were made with a biogeochemical model coupled to a global ocean general circulation model (OGCM), which itself was forced by 55-year NCEP reanalysis fields. In the first simulation, atmospheric CO₂ was maintained at the preindustrial level (278 ppmv); in the second simulation, atmospheric CO₂ followed the observed increase. Simulated air-sea CO₂ fluxes and associated variables were then analysed with a statistical tool known as multichannel singular spectrum analysis (MSSA). We found that the subtropical gyre is not the largest contributor to the overall, basin-wide variability, in contrast to previous suggestions. The subpolar gyre and the inter-gyre region (the transition area between subpolar and subtropical gyres) also contribute with multipolar anomalies at multiple frequencies: these tend to cancel one another in terms of the basin-wide air-sea CO₂ flux. We found a strong correlation between the air-sea CO₂ fluxes and the North Atlantic Oscillation (NAO), but only if one takes into account time lags as does MSSA (maximum r=0.64 for lags between 1 and 3 years). The effect of increasing atmospheric CO₂ (the anthropogenic perturbation) on total variability was negligible at interannual time scales, whereas at the decadal (13-year) time scale, it increased variability by 30%.