It is generally assumed that the overall results of a clinical trial are generalisable to all patients in the trial and all similar future patients; in other words, that the relative treatment effect in individual patients is similar to the overall trial result. Although this assumption underpins the application of trial results to clinical practice, it has rarely been tested. By independently derived prognostic models, the results of the European Carotid Surgery Trial and the UK-TIA Aspirin Trial were reanalysed to find out whether relative treatment effect varied with absolute baseline risk of stroke. There was significant heterogeneity of relative treatment effect in both trials, resulting in substantial variation in absolute treatment effect with predicted baseline risk. Although, on average, the application of overall trial results to all patients will do more good than harm, a knowledge of the association between relative treatment effect and absolute baseline risk will increase the cost-effectiveness of healthcare interventions by identifying those patients in whom treatment is ineffective and those patients who are most likely to benefit.