Adaptive designs are one of the most promising developments in statistics with applications to clinical trials. Obviously, knowledge at the beginning of a clinical trial will always be limited. Thus it may seem logical that the knowledge from accumulating information should be used to optimize the design of the trial. This paper discusses conditions under which this may be possible in phase III clinical trials where the principal aim is confirmation of hypotheses that have been developed in earlier stages of drug development.