CO2 exchange between lake water and the atmosphere was investigated at Toolik Lake (Alaska) and Soppensee (Switzerland) employing the eddy covariance (EC) method. The results obtained from three field campaigns at the two sites indicate the importance of convection in the lake in driving gas flux across the water-air interface. Measurements were performed during short (1--3 day) periods with observed diurnal changes between stratified and convective conditions in the lakes. Over Toolik Lake the EC net CO2 efflux was 114 ¿ 33 mg C m-2 d-1, which compares well with the 131 ¿ 2 mg C m-2 d-1 estimated by a boundary layer model (BLM) and the 153 ¿ 3 mg C m-2 d-1 obtained with a surface renewal model (SRM). Floating chamber measurements, however, indicated a net efflux of 365 ¿ 61 mg C m-2 d-1, which is more than double the EC fluxes measured at the corresponding times (150 ¿ 78 mg C m-2 d-1). The differences between continous (EC, SRM, and BLM) and episodic (chamber) flux determination indicate that the chamber measurements might be biased depending on the chosen sampling interval. Significantly smaller fluxes (p < 0.06) were found during stratified periods (51 ¿ 42 mg C m-2 d-1) than were found during convective periods (150 ¿ 45 mg C m-2 d-1) by the EC method, but not by the BLM. However, the congruence between average values obtained by the models and EC supports the use of both methods, but EC measurements and the SRM provide more insight into the physical-biological processes affecting gas flux. Over Soppensee, the daily net efflux from the lake was 289 ¿ 153 mg C m-2 d-1 during the measuring period. Flux differences were significant (p < 0.002) between stratified periods (240 ¿ 82 mg C m-2 d-1) and periods with penetrative convection (1117 ¿ 236 mg C m-2 d-1) but insignificant if convection in the lake was weak and nonpenetrative. Our data indicate the importance of periods of heat loss and convective mixing to the process of gas exchange across the water surface, and calculations of gas transfer velocity using the surface renewal model support our observations. Future studies should employ the EC method in order to obtain essential data for process-scale investigations. Measurements should be extended to cover the full season from thaw to freeze, thereby integrating data over stratified and convective periods. Thus the statistical confidence in the seasonal budgets of CO2 and other trace gases that are exchanged across lake surfaces could be increased considerably.