If equivalence of two treatments has to be proved by an experiment wit
h normally distributed observations and fixed sample size, there are t
wo usual methods. On the one hand, there is the double t test and, on
the other hand, a test procedure suggested by Anderson and Hauck, whic
h is uniformly more powerful than the first one but only capable of ke
eping asymptotically to the nominal level. Based on these two methods
a new test is presented, which is uniformly more powerful than the dou
ble t test and also keeps the preassigned level.