This presentation is motivated by the experience of a subgroup analysis of an orphan drug. Analyses within subgroups are common in clinical studies but subgroups based on post-randomization measures should be avoided. Particularly, subgroup analysis by response status after randomization could lead to improper inference if not conducted and interpreted cautiously. Benefits and limitations of such subgroup analysis will be discussed. Examples of such subgroup analysis in small clinical trials will be given. Statistical methods and outcomes will be presented.