OBJECTIVE: In its original format SF-36 is a profile measure that results in 2 or more subscale scores. This design means that SF-36 cannot be used in cost-effectiveness analysis in which health outcomes must be represented as a single summary index of the type exemplified by EQ-5Dindex. However, many clinical studies continue to report outcomes in terms of SF-36 and the researchers continue to confront the challenge of “salvaging” such data for use in economic analysis. This paper reviews the dangers in using SF-36/EQ-5D conversion models that have been (inappropriately) calibrated using UK social preferences weights. METHODS: The Franks (2004) and Lawrence (2004) algorithms for computing EQ-5Dindex values from SF-36 responses were applied to data collected in a UK national population survey (n ~ 16,000) in which respondents had also completed an EQ-5D questionnaire. The estimated and observed EQ-5Dindex scores were computed for 6 major diagnostic groups (respiratory, endocrine, circulatory, musculoskeletal, cancer, mental health). The published US scoring algorithm for EQ-5Dindex was also applied to these responses. The potential 1-year QALY gains for each group were estimated using UK population age/sex norms for EQ-5D, assuming complete remission of current ill-health condition. RESULTS: Although for some diagnostic groups mean absolute differences appeared relatively small, differences between estimated and observed EQ-5Dindex varied enormously. QALY gains varied when EQ-5Dindex scores were replaced by Franks and Lawrence's scores. The rank order of these gains was not significantly altered when US EQ-5D weights were used in place of UK social preferences. CONCLUSION: Conversion models that directly estimate EQ-5Dindex scores are dangerously vulnerable to the value set used to estimate the regression model. Models calibrated using UK population preference weights should only be in extremis and with due care where domestic national values re unavailable.