OBJECTIVES: To review the health economic literature for drugs that have since been withdrawn for safety reasons. To assess the cost-effectiveness of the non-sedating anti-histamine, terfenadine (withdrawn in 1998 for its cardiotoxicity) compared with chlorpheniramine, a sedating anti-histamine which is still available. METHODS: Literature searches of economic evaluations of drug that have been withdrawn since 1980 were conducted. A decision analysis comparing terfenadine with chlorpheniramine was developed. Observational data were obtained for estimates of the incidence of ventricular dysrythmias, the incidence of serious injuries as a consequence of sedation with chlorpheniramine, and the risk of death resulting from arrhythmias or serious injuries. Health state utilities and data on clinical effectiveness were obtained from published sources. A 12-month time horizon was adopted, with probabilistic sensitivity analysis and threshold analyses conduced to assess the impact of uncertainty in parameter estimates. RESULTS: Examples of published cost-effectiveness analyses included troglitazone (PPAR activator, withdrawn in 2000 for hepatotoxicity), and rofecoxib (COX-2 inhibitor, withdrawn in 2004 for cardiovascular toxicity). Despite including ADRs in the analyses, both drugs were deemed cost-effective. The analysis of treatments for allergic rhinitis revealed that, in fact, chlorpheniramine had a less favourable risk/benefit ratio than terfenadine, with a mean difference of 3.5 QALYs per 1000 patients (95% credible interval, 0.3, 7.6). Threshold analysis suggested that it would require the relative risk of serious injury with terfenadine, compared with chlorpheniramine, to increase from 45% to 85%, or for the efficacy of terfenadine to reduce from 60% to 34% for the decision to be reversed. CONCLUSION: The inclusion in economic evaluations of ADRs that are deemed too hazardous to warrant market authorisation by regulators, may lead to counter-intuitive estimates of cost-effectiveness. This may be the fault of regulators for not adopting decision analytic models, or reflect a lack of risk aversion in economic evaluations. Alternative explanations are explored.