In the 'total evidence' approach to phylogenetics, the reliability of a clade is implicitly measured by its degree of support, often embodied in a robustness index such as a bootstrap proportion. In the taxonomic congruence approach, the measurement of reliability has been implemented by various consensus or supertree methods, but was seldom explicitly discussed as such. We explore a reliability index for clades using their repetition across independent data sets. All possible combinations of the elementary data sets are used to compose the sets of independent data sets, across which the repetitions are counted. The more a clade occurs across such independent combinations, the higher its index. However, if other repeated clades occur that are incompatible with that clade, its index is decreased to take into account the uncertainty resulting from conflicting hypotheses. Results can be summarized through a greedy consensus tree in which clades appear according to their repetition indices. This index is tested on a 73 acanthomorph taxa data set composed of five independent molecular markers and multiple combinations of them. On this particular application, we confirm that reliability as defined here and robustness (estimated by bootstrap proportions obtained from a 'total evidence' approach) should be clearly distinguished. Corresponding author: Guillaume Lecointre, Équipe 'Phylogénie', UMR 7138 'Systématique, Adaptation, Évolution', Museum National d'Histoire Naturelle, Département Systématique et Évolution, case postale 26, 57 rue Cuvier, 75231 Paris cedex 05, France.