Net ecosystem exchange (NEE) was estimated for an area of tundra near Abisko using both eddy covariance (EC) data and chamber measurements. This area of tundra is heterogeneous with six principal elements forming a landscape mosaic. Chamber measurements in patches of the individual mosaic elements were used to model NEE as a function of irradiance and temperature. The area around the EC mast was mapped, and a footprint model was used to simulate the varying source fraction attributable to each mosaic element. Various upscaling approaches were used to estimate NEE for comparison with NEE calculated from the EC observations. The results showed that EC measurements made for such a heterogeneous site are robust to the variations in NEE between mosaic elements that also vary substantially in their source fractions. However, they also revealed a large (?60%) bias in the absolute magnitude of the cumulative negative NEE for a 40-day study period simulated by various upscaling approaches when compared to the value calculated from the EC observations. The magnitude of this bias, if applied to estimates for the entire tundra region, is substantial in relation to other components of the global carbon budget. Various hypotheses to account for this bias are discussed and, where possible, evaluated. A need is identified for more systematic sampling strategies when performing chamber measurements in order to assess the extent to which subjectivity of chamber location may account for much of the observed bias. If this is the origin of the bias, then upscaling approaches using chamber measurements may generally overestimate CO2 uptake.