The application of a classification technique, called numerical taxonomy (NT) to the selection of a set of preferred liquid phases is discussed. The liquid phases are considered to be a population of individuals which must be fitted in a hierarchical classification, comparable to the systematic classification of species in biology. This is achieved by using McReynolds' data (J. Chromatog. Sci. 8, 685 (1970)) to calculate a coefficient of resemblance between phases, called the distance. With these distances a resemblance matrix is constructed. The weighted pair group classification method is then used to derive the hierarchical classification from this matrix. The set of n preferred liquid phases can then be obtained by choosing from each of the n most dissimilar classes, one phase which responds best to other criteria such as availability and stability. A comparison of the results obtained is carried out with results obtained by other authors using information theory and principal components analysis. It is concluded that a good agreement is obtained and that numerical taxonomy has some advantages compared with the other techniques.