Global concern over rising atmospheric CO(2) concentrations has led to a proliferation of studies of carbon cycling in terrestrial ecosystems. Associated with this has been widespread adoption of the eddy covariance method to provide direct estimates of mass and energy exchange between vegetation surfaces and the atmosphere. With the eddy covariance method, fast-response instruments ( 10 - 20 Hz) are typically mounted above plant canopies and the fluxes are calculated by correlating turbulent fluctuations in vertical velocity with fluctuations in various scalars such as CO(2), water vapour and temperature. These techniques allow the direct and non-destructive measurement of the net exchange of CO(2) owing to uptake via photosynthesis and loss owing to respiration, evapotranspiration and sensible heat. Eddy covariance measurements have a high temporal resolution, with fluxes typically calculated at 30-min intervals and can provide daily, monthly or annual estimates of carbon uptake or loss from ecosystems. Such measurements provide a bridge between 'bottom-up' ( e. g. leaf, soil and whole plant measures of carbon fluxes) and 'top-down' approaches ( e. g. satellite remote sensing, air sampling networks, inverse numerical methods) to understanding carbon cycling. Eddy covariance data also provide key measurements to calibrate and validate canopy- and regional-scale carbon balance models. Limitations of the method include high establishment costs, site requirements of. at and relatively uniform vegetation and problems estimating fluxes accurately at low wind speeds. Advantages include spatial averaging over 10 - 100 ha and near-continuous measurements. The utility of the method is illustrated in current flux studies at ideal sites in northern Australia. Flux measurements spanning 3 years have been made at a mesic savanna site ( Howard Springs, Northern Territory) and semi-arid savanna ( Virginia Park, northern Queensland). Patterns of CO(2) and water vapour exchange at diurnal, seasonal and annual scales are detailed. Carbon dynamics at these sites are significantly different and reflect differences in climate and land management ( impacts of frequent. re and grazing). Such studies illustrate the utility of the eddy covariance method and its potential to provide accurate data for carbon accounting purposes. If full carbon accounting is implemented, for ideal sites, the eddy covariance method provides annual estimates of carbon sink strength accurate to within 10%. The impact of land-use change on carbon sink strength could be monitored on a long-term basis and provide a valuable validation tool if carbon trading schemes were implemented.