A note on multiple testing procedures in dose finding.

It is shown that, in some recently proposed multiple test procedures for dose finding (Tamhane, Hochberg, and Dunnett, 1996, Biometrics 52, 21-37), strong control of the familywise error is only guaranteed if order relations are assumed to hold among the means. Procedures that can be applied safely without assuming any restrictions on the means can be based on the classical many--one comparisons of doses with the zero dose.