OBJECTIVES: Decisions about the use of new medical technologies based on estimates of the average cost-effectiveness across a potentially heterogeneous population runs the risk of foregoing net health benefits(NHB) for sub-groups of the population. We propose a general framework within which to assess between-patient heterogeneity and its role in cost-effectiveness subgroup analysis(CESA), complementing this with a practical application.  METHODS: We first describe how to extend methods for cost-effectiveness analysis (based on current information) to address issues such as estimation of NHB, sources of heterogeneity, definition and selection of subgroups.  Next, we define the role of uncertainty in CESA, extending the concept of Value of Information(VoI) to include the notion of a static and dynamic Value of Heterogeneity(VoH).  The application of the proposed theoretical framework is illustrated using a cost-effectiveness model developed for the analysis of a multicentre-trial(RITA-3), which assessed the efficacy of an early arteriography with revascularisation versus standard management in patients with acute coronary-syndrome.  Using this model we conducted a re-analysis investigating alternative subgroup specifications, varying between one and five subgroups, with a view to produce an efficiency frontier for subgroup analysis relating to this decision problem.  We assessed the static and dynamic VoH under each specification. RESULTS: The population expected NHB when considering five subgroups was 105,500 QALYs greater than decision based on estimates for the average population (static-VoH). Although, identifying 5 subgroups reduced in the Expected Value of Perfect Information(EVPI) (920 QALYs, at a threshold of £30,000/QALY), the potential NHB from resolving uncertainty was greater after heterogeneity has been identified (dynamic-VoH). CONCLUSIONS: Our initial findings support the argument that explicit consideration of heterogeneity in CEA leads to a positive static and dynamic VoH. In addition, heterogeneity not only may increase the EVPI but can also reduce its magnitude. The VoH framework offers a useful guidance for a more systematic CESA.