The evaluation of rehabilitation programmes may be distorted by regression to the mean: In a group of patients with extreme measurement values, these values tend to be less extreme on a following point in time due to merely random components and regardless of a "true" treatment effect. If this effect is not taken into account the effectiveness of rehabilitation programmes may be estimated wrongly. In this paper regression to the mean is explained comprehensively, and common misunderstandings are clarified. It is shown which conditions are crucial for occurrence of regression to the mean and which factors determine its strength. Furthermore it is shown how regression to the mean should be controlled in repeated measurement designs.