Lake-atmosphere CO2 flux was directly measured above a small, woodland lake using the eddy covariance technique and compared with fluxes deduced from changes in measured lake-water CO2 storage and with flux predictions from boundary-layer and surface-renewal models. Over a 3-yr period, lake-atmosphere exchanges of CO2 were measured over 5 weeks in spring, summer, and fall. Observed springtime CO2 efflux was large (2.3-2.7 ??mol m-2 s-1) immediately after lake-thaw. That efflux decreased exponentially with time to less than 0.2 ??mol m-2 s-1 within 2 weeks. Substantial interannual variability was found in the magnitudes of springtime efflux, surface water CO2 concentrations, lake CO2 storage, and meteorological conditions. Summertime measurements show a weak diurnal trend with a small average downward flux (-0.17 ??mol m-2 s-1) to the lake's surface, while late fall flux was trendless and smaller (-0.0021 ??mol m-2 s-1). Large springtime efflux afforded an opportunity to make direct measurement of lake-atmosphere fluxes well above the detection limits of eddy covariance instruments, facilitating the testing of different gas flux methodologies and air-water gas-transfer models. Although there was an overall agreement in fluxes determined by eddy covariance and those calculated from lake-water storage change in CO2, agreement was inconsistent between eddy covariance flux measurements and fluxes predicted by boundary-layer and surface-renewal models. Comparison of measured and modeled transfer velocities for CO2, along with measured and modeled cumulative CO2 flux, indicates that in most instances the surface-renewal model underpredicts actual flux. Greater underestimates were found with comparisons involving homogeneous boundary-layer models. No physical mechanism responsible for the inconsistencies was identified by analyzing coincidentally measured environmental variables.