The air-sea transfer velocity of CO2, kCO2, from a deliberate tracer experiment (DTE) and from eddy covariance measurements (ECM) during ASGAMAGE differed on average by a factor of ~2.5, in spite of the fact that both methods were applied with maximum care. In an attempt to explain this difference, implicit assumptions were investigated critically using complementary one-dimensional (1-D) and 2-D models for the upper oceanic boundary layer and atmospheric surface layer. Simulations of a DTE and ECM facilitate examination of the possible impact of near-surface vertical gradients, horizontal inhomogeneity, and instationarity of the tracer concentration on the results from field experiments. The simulations suggest that vertical tracer concentration gradients due to physical processes might occur outside the diffusive boundary and that these might have escaped prior observation due to their confinement to the upper meter of the water column. Such gradients could then cause DTE to typically underestimate kCO2 by 10--25%, depending on ambient conditions, such as light penetration. Buffering of the seawater CO2 system renders ECM much less sensitive to near-surface gradients. However, inhomogeneity of the CO2 concentration in water might cause scatter in ECM due to inhomogeneity-related flux divergence in air. Additionally, in the case of instationarity a complex and strongly varying fine structure of vertical CO2 concentration profiles in water could cause inconsistencies with CO2 flux observations in air. Both mechanisms might explain several tens of percents of the scatter in the ECM, while systematic effects are ~10%.