Procedures are currently available for the evaluation of hierarchical classifications of produce tree dissimilarities or consensus dendrograms. Some tests of cluster validity operate by comparing all possible partitions from a tree with a reference partition. We propose an exhaustive search procedure to compare all partitions from one dendrogram with all partitions derived from the other to detect hierarchical levels at which the two dendrograms show maximum agreement. The method is illustrated using RAPD and microsatellite data in order to detect clones in reed populations. The utility of our approach is its ability to reveal extra information in different genetic data sets which would be hidden otherwise. The method is also useful in any field of science where hierarchical clustering is the main research tool and comparison of results is an objective. Artificial and actual dendrograms, together with randomly simulated trees were used to compare the performance of five classical coefficients of partition dissimilarity. The simulations showed that when meaningful group structure is lacking, then the five coefficients are in full disagreement, but they perform similarly otherwise.