In a recent paper, Brockho€ and Schlich (1998, Handling replications in discrimination tests. Food Quality and Preference, 9(5), 303±312) proposed a statistically sound way of handling replications in di€erence testing. In the present paper, this new test is applied to the data obtained in six experiments on non alcoholic beverages, where triangle tests were intensively replicated (between eight and 60 times) with groups of subjects composed of 12±61 students. The paper aims to estimate in these practical situations the extent to which a group of panelists is heterogeneous towards the ability of detecting a sensory di€erence among two products. As the results indicate that group heterogeneity was lower than Brockho€ and Schlich ®rst guessed, the value of using replicates in di€erence testing is emphasized. It seems that, average over experiments, at least 10 replicates were necessary to properly estimate the level of heterogeneity and that a single subject having done 10 replicates would bring to our test the same amount of informa- tion as ®ve subjects having done the test once only.