The eddy covariance technique is widely used to measure CO2 and other gas fluxes. However, eddy covariance fluxes are affected by systematic errors that must be corrected. Among them, high frequency loss corrections are particularly important in this regard, especially when using a closed-path infrared gas analyzer. In this study, we compared three approaches to do these corrections for CO2 fluxes and evaluated their impact on the carbon balance an intensively grazed grassland site in Belgium . In the first approach, the computation of correction factors was based on the measured sensible heat cospectra (‘local’ cospectra), whereas the other two were based on theoretical models (Kaimal et al., 1972). The correction approaches were validated by comparing the nighttime eddy covariance CO2 fluxes corrected with each approach and in situ soil respiration measurements. We found that the local cospectra differed from the Kaimal theoretical shape, although the site could not be considered ‘difficult’ (i.e., fairly flat, homogeneous, low vegetation, sufficient measurement height), appearing less peaked in the inertial subrange. This difference greatly affected the correction factor, especially for night fluxes. Night fluxes measured by eddy covariance were found to be in good accord with in situ soil respiration measurements when corrected with local cospectra and to be overestimated when corrected with Kaimal cospectra. As the difference between correction factors was larger in stable than unstable conditions, this acts as a selective systematic error and has an important impact on annual fluxes. On the basis of a 4-year average, at DTO the errors reach 71-150 g C m-2 y-1 for net ecosystem exchange (NEE), 280-562 g C m-2 y-1 for total ecosystem respiration (TER) and 209-412 g C m-2 y-1 for gross primary productivity (GPP), depending on the approach used. We finally encourage site PIs to check the cospectrum shape at their sites and, if necessary, compute frequency correction factors on the basis of local cospectra ...