There has been an increasing need to measure the exchange of CO2 between the atmosphere and vegetated surfaces for large areas in order to quantify the carbon budget of the terrestrial biosphere. The boreal landscape is heterogeneous owing to different forest cover types and disturbance regimes, and regional quantification of CO2 flux is difficult without numerous species-specific flux measurements. During the Boreal Ecosystem-Atmosphere Study in 1994 and 1996 the National Research Council of Canada operated a Twin Otter aircraft that measured CO2, sensible and latent heat fluxes, and other trace gases over boreal forests in Saskatchewan, Canada. A flux-unmixing method was developed to calculate flux densities for eight major cover types from the aircraft-based measurements. Using a coregistered land cover map at 30-m resolution derived from Landsat thematic mapper data, the contribution of each cover type to the CO2 flux measured by the aircraft was estimated using a contributing area (footprint) function according to the wind direction, the atmospheric stability, the horizontal distance of each pixel from the aircraft, and aircraft height. The unmixing method uses a linear inversion method with the footprint-weighted cover type fractions as the set coefficients for each segment of a flight line. In the inversion, various constraint strategies were used to confine the inversion results to minimize the effect of various sampling errors. It is shown that (1) mathematical constraint is critically important in the inversion, (2) a simple constraint toward the mean flux values is effective in producing reasonable inversion results, and (3) the inversion accuracy can be further improved when simultaneous tower measurements in the dominant cover types are used as tight constraints. With such constraints the estimated fluxes from the cover types without tower measurements appear to be reasonable. It is concluded that aircraft measurement adds to our ability to map the regional flux field using remote sensing images because (1) it allows the derivation of flux data for cover types without tower-based measurements and (2) it can be used to infer the representativeness of tower measurements for the measured cover types in the landscape. ¿ 1999 American Geophysical Union