Analyses of disjoint subsets of study subjects, based on demographic or baseline characteristics, is common in the analysis of clinical studies. We review potential objectives of subgroup analyses. These may include a general data dredge to look for signals that some subset of patients does not benefit from a treatment, a confirmation that a prespecified subset of concern does benefit from a treatment, investigation of similarity of magnitude of benefit, or other reason. Inferential testing is common, either to assess interactions in disjoint subgroups or to assess benefit within a subgroup. We are skeptical of such testing. Rather, we prefer descriptive (especially graphical) procedures for crude screening of numerous subgroups. Finally, assessment of risk in subgroups is considered so that the risk-benefit profile can be considered within and between subgroups.