Two-step tests based on historical control data are proposed for detecting
different means when experimental and control populations are normally dist
ributed with possibly different known variances. The results readily extend
to the case where the control and experimental data are dichotomous. Provi
sions for early acceptance and early rejection are included. Control Clin T
rials 1999;20:353-364 (C) Elsevier Science Inc. 1999.