It has often been found desirable, to strengthen palaeoecological inferences, to be able to separate mixtures of pollen taxa whose morphological characteristics overlap considerably, as with the species of many circumboreal tree genera (birch and pine being much-studied examples). Statistical methods are potentially valuable in this problematic field. Previously developed numerical approaches are discussed. Methods based on the identification of single grains are shown to be less satisfactory than methods involving the fitting of frequency histograms, corresponding to possible combinations of known reference distributions, to the observed frequency histograms from fossil material, as proposed by Eneroth (1951). A solution of this type based on the maximum likelihood principle is described and applied to some of Eneroth's own data. The assumptions and limitations of the method, particularly with regard to size statistics versus other morphological data, are discussed.