The most direct method for flux estimation uses eddy covariance, which is also the most commonly used method for land-based measurements of surface fluxes. Moving platforms are frequently used to make measurements over the sea, in which case motion can disturb the measurements. An alternative method for flux estimation should be considered if the effects of platform motion cannot be properly corrected for. Three methods for estimating CO2 fluxes are studied here: the eddy-covariance, the inertial-dissipation, and the cospectral-peak methods. High-frequency measurements made at the land-based Östergarnsholm marine station in the Baltic Sea and measurements made from a ship during the Galathea 3 expedition are used. The Kolmogorov constant for CO2, used in the inertial-dissipation method, is estimated to be 0.68 and is determined using direct flux measurements made at the Östergarnsholm site. The cospectral-peak method, originally developed for neutral stratification, is modified to be applicable in all stratifications. With these modifications, the CO2 fluxes estimated using the three methods agree well. Using data from the Östergarnsholm site, the mean absolute error between the eddy-covariance and inertial-dissipation methods is 0.25 μmol m−2 s−1. The corresponding mean absolute error between the eddy-covariance and cospectral-peak methods is 0.26 μmol m−2 s−1, while between the inertial-dissipation and cospectral-peak methods it is 0.14 μmol m−2 s−1.