AbstractRelative to the problem presented and treated by Parisot in an earlier issue of this journal,1we present a new solution based on our approach of hierarchical classification. The methodology used, that we will describe in a concise way, allows us to organize either the set of the descriptive variables or the set of the individuals. Its extreme generality includes a set theoretic coding which respects faithfully the globality of the intrinsic nature of the descriptive variables. Very rich results are obtained in terms of classification trees reduced to their ‘significant’ levels. One of the most significant levels corresponds to the partition obtained by Parisot. The ‘significant’ nodes indicate the most relevant associations. The signification is conceived from the empirical distribution — on the successive tree levels — of association coefficients between partition and ‘preordonnance’ (i.e. total preorder on the set of unordere