Difficulty in balancing the global carbon budget has lead to increased attention on tropical forests, which have been estimated to account for up to one third of global gross primary production. Whether tropical forests are sources, sinks, or neutral with respect to their carbon balance with the atmosphere remains unclear. To address this issue, estimates of net ecosystem exchange of carbon (NEE) were made for 3 years (1998–2000) using the eddy-covariance technique in a tropical wet forest in Costa Rica. Measurements were made from a 42 m tower centred in an old-growth forest. Under unstable conditions, the measurement height was at least twice the estimated zeroplane height from the ground. The canopy at the site is extremely rough; under unstable conditions the median aerodynamic roughness length ranged from 2.4 to 3.6 m. No relationship between NEE and friction velocity (u*) was found using all of the 30-min averages. However, there was a linear relationship between the nighttime NEE and averaged u* (R2 = 0.98). The diurnal pattern of flux was similar to that found in other tropical forests, with mean daytime NEE ca. − 18 μmol CO2 m−2 s−1 and mean nighttime NEE 4.6 μmol CO2 m−2 s−1. However, because ∼ 80% of the nighttime data in this forest were collected during low u* conditions (< 0.2 m s−1), nighttime NEE was likely underestimated. Using an alternative analysis, mean nighttime NEE increased to 7.05 μmol CO2 m−2 s−1. There were interannual differences in NEE, but seasonal differences were not apparent. Irradiance accounted for ∼ 51% of the variation in the daytime fluxes, with temperature and vapour pressure deficit together accounting for another ∼ 20%. Light compensation points ranged from 100 to 207 μmol PPFD m−2 s−1. No was relationship was found between 30-min nighttime NEE and tower-top air temperature. A weak relationship was found between hourly nighttime NEE and canopy air temperature using data averaged hourly over the entire sampling period (Q10 = 1.79, R2 = 0.17). The contribution of below-sensor storage was fairly constant from day to day. Our data indicate that this forest was a slight carbon source in 1998 (0.05 to −1.33 t C ha−1 yr−1), a moderate sink in 1999 (−1.53 to −3.14 t C ha−1 yr−1), and a strong sink in 2000 (−5.97 to −7.92 t C ha−1 yr−1). This trend is interpreted as relating to the dissipation of warm-phase El Niño effects over the course of this study.