A system has been developed to directly measure isoprene flux above a forest canopy by eddy covariance using the combination of a fast response, real-time isoprene sensor and sonic anemometer. This system is suitable for making nearly unattended, long-term and continuous measurements of isoprene fluxes. Isoprene detection is based on chemiluminescence between isoprene and reactant ozone, which produces green light at 500 nm. The sensor has a noise level (1σ) of 450 pptv for a 1-s integration which is dominated by random high-frequency noise that does not significantly degrade eddy covariance flux measurements. Interference from the flux of other compounds is primarily due to the emission of monoterpenes, propene, ethene and methyl butenol and the deposition of methacrolein and methyl vinyl ketone. The average total interference for North American landscapes in midday summer is estimated to be about 5% for emissions and -3% for deposition fluxes. In only a few North American landscapes, where isoprene emissions are very low and methyl butenol emissions are high, are interferences predicted to be significant. The system was field tested on a tower above a mixed deciduous forest canopy (Duke Forest, North Carolina, USA), dominated by oak trees, which are strong isoprene emitters. Isoprene fluxes were estimated for 307 half-hour sampling periods over 10 days. Daytime fluxes ranging from 1 to 14 mg C m-2 h-1 were strongly correlated with light and temperature. The daytime mean flux of 6 mg C m-2 h-1 were strongly correlated with light and temperature. The daytime mean flux of 6 mg C m-2 h-1 is similar to previous estimates determined by relaxed eddy accumulation by Geron et al.  at this site. Nighttime fluxes were near zero (0.01 ± 0.03 mg C m-2 h-1).