My fellow editors of ATLA, Oliver Flint (USA) and Michael Balls (UK), have contributed to the opinions on in vitro test validation in editorials published in the last two issues of the journal. 1 • 2 These editorials have been centred around the recommendations of the Amden Report, 3 Flint critisising what he sees as a lack of emphasis on mechanistic considerations, and Balls restating and reinforcing the conclusions of the Amden Workshop.Flint suggests that there should be one overriding rule for validation: "All tests must have demonstrable mechanistic similarity to the events we wish to study in vivo" (Flint's Law; Balls' denomination), whereas Balls' opinion is that the criteria for validation must be adapted to the type of validation under consideration (Balls' uncertainty principle; my denomination). When one reads the two compositions together, one finds that there is not such a tremendous difference between them. Both Flint and Balls are deeply concerned by questions of how best to develop test procedures, how to validate them correctly and how to formulate a maxim for validation studies. They are certainly both entitled to their beliefs, and I am not going to take sides in any way, other than by explaining some of my own beliefs.Validation should be performed as a research project. Consequently, the purpose of validation is to test one or several hypotheses, formulated on the basis of results obtained from basic cell biological studies, from mechanistic toxicological studies or in toxicity test development. The validity of the results of the project in relation to each hypothesis should then be judged by means of statistical procedures. Some of the ideas investigated in the MEIC-programme 4 may serve as examples of relevant hypotheses: