One approach to delivering cost-effective healthcare requires the identification of patients as individuals or subpopulations that are more likely to respond to an appropriate dose and/or schedule of a therapeutic agent, or as subpopulations that are less likely to develop an adverse event (i.e., personalized or stratified medicine). Biomarkers that identify therapeutically relevant variations in human biology are often only uncovered in the later stage of drug development. In this article, the Industry Pharmacogenomics Working Group provides, for regulatory consideration, its perspective on the rationale for the conduct of what is commonly referred to as the prospective-retrospective analysis (PRA) of biomarkers. Reflecting on published proposals and materials presented by the US FDA, a decision tree for generating robust scientific data from samples collected from an already conducted trial to allow PRA is presented. The primary utility of the PRA is to define a process that provides robust scientific evidence for decision-making in situations where it is not necessary, nor practical or ethical to conduct a new prospective clinical study.