Technical Abstract: We introduce a new eddy-covariance method that uses a spectral decomposition algorithm called empirical mode decomposition. The technique is able to calculate contributions to near-surface fluxes from different periodic components. Unlike traditional Fourier methods, this method allows for non-orthogonal contributions to the total flux, which are shown to be errors due to the undersampling of low-frequency processes. Inspection of the non-orthogonal terms with relation to sampling duration and periodicity reveals that a measured periodic process requires approximately six cycles in order to be sufficiently sampled. This determines the maximum eddy size sufficiently measured given a particular sampling duration.