The classification of Raoulia and allied genera in New Zealand is reviewed with the aid of numerical phenetic analysis. Two analyses were carried out, one on 117 specimens of Raoulia using 84 morphological characters and the other on 79 specimens of Gnaphaliinae using 165 morphological characters. At least seven species exist, all falling within the currently accepted limits of Raoulia, additional to those recognised by Allan. The genus as currently recognised is heterogeneous, containing three distinct groups and several isolated species. The present subgeneric and sectional arrangement of the species is demonstrated to be inappropriate. The bulk of Raoulia should be split into two genera. The affinities of R. cinerea lie with Helichrysum. Generic limits elsewhere in the group are shown to be inaccurately defined. Thus Helichrysum bellidioides should be transferred to Anaphalioides (Benth.) Kirp., Leucogenes has closest ties to Raoulia subg. Psychrophyton and the New Zealand and Australian species of Ewartia are not congeneric. Numerical phenetic analysis is shown to be a useful tool in classification, particularly in a group such as the Gnapha1iinae in which genera are not clearly defined. Experiments in varying different components of the data base and of the numerical method produce many different arrangements of the same plant group, and demonstrate that no one arrangement is of necessity a reliable indication of phenetic relationships and, furthermore, that different results may all provide information useful in classification. Analyses based on floral and vegetative characters produced results different from each other and from the two combined; floral characters contribute more heavily to the overall arrangement than do vegetative ones and their contribution is greater at higher taxonomic levels. Altering the similarity measure from Gower's general coefficient of similarity to the simple matching coefficient has little effect, as does removing the weighting effect of additive coding of multi-state characters. In comparison, the effect of varying the clustering technique is marked, with the results of complete linkage and especially centroid sorting proving low in congruence with the similarity matrix while single linkage produces a phenograrn with low structural value. In spite of the presence of markedly unequal samples of different taxa, very little difference is found between UPGMA and WPGMA.