A clustering method to reduce the difficulties in selection found in plant trials by the frequently existing qualitative or crossover interaction is described. It is based on an order-constrained inferiority score and avoids the need to postulate a model for the interaction. The resulting dendrogram is analyzed to arrive at a proper stopping point. Each final cluster includes a 'leader' genotype, which has the lowest inferiority score and would show no significant crossover (rank change) with the others. The method can be seen as preselection of some genotypes, each one uniformly not inferior to others in the same group; the latter can be disregarded for subsequent analysis. When the procedure is compared with some previous proposals it seems to provide a better and simpler solution.