A comprehensive evaluation of energy balance closure is performed across 22 sites and 50 site-years in FLUXNET, a network of eddy covariance sites measuring long-term carbon and energy fluxes in contrasting ecosystems and climates. Energy balance closure was evaluated by statistical regression of turbulent energy fluxes (sensible and latent heat (LE)) against available energy (net radiation, less the energy stored) and by solving for the energy balance ratio, the ratio of turbulent energy fluxes to available energy. These methods indicate a general lack of closure at most sites, with a mean imbalance in the order of 20%. The imbalance was prevalent in all measured vegetation types and in climates ranging from Mediterranean to temperate and arctic. There were no clear differences between sites using open and closed path infrared gas analyzers. At a majority of sites closure improved with turbulent intensity (friction velocity), but lack of total closure was still prevalent under most conditions. The imbalance was greatest during nocturnal periods. The results suggest that estimates of the scalar turbulent fluxes of sensible and LE are underestimated and/or that available energy is overestimated. The implications on interpreting long-term CO2 fluxes at FLUXNET sites depends on whether the imbalance results primarily from general errors associated with the eddy covariance technique or from errors in calculating the available energy terms. Although it was not entirely possible to critically evaluate all the possible sources of the imbalance, circumstantial evidence suggested a link between the imbalance and CO2 fluxes. For a given value of photosynthetically active radiation, the magnitude of CO2 uptake was less when the energy imbalance was greater. Similarly, respiration (estimated by nocturnal CO2 release to the atmosphere) was significantly less when the energy imbalance was greater.