Usual statistical practice in inference on the comparison of two treatment effects is based on significance tests. In the positive case, it is unlikely that the observed difference is based only on random variation, and a decision is made that there is a real superiority. In the negative case, no conclusion on a missing difference is possible. In both cases, statistical tests fail to describe the magnitude of the underlying effect. Confidence intervals are more suitable for demonstrating clinical efficacy. Their advantages and applications are discussed.