Some cautionary comments on the use of retrospective designs to evaluate treatment efficacy.

This article contrasts the true experimental approach with nonexperimental alternatives, such as retrospective designs, for research on treatment efficacy. The potential problems and limitations of the nonexperimental approach are highlighted so that when deviations from the more desirable experimental methodology are necessary, they can be undertaken with full knowledge of the hazards to be expected. Important issues are illustrated through discussion of a specific type of retrospective design, the case-control design.