OBJECTIVES: It is commonly assumed that the gold standard method for producing new language versions of patient-completed outcome measures is to translate using back-translation. This paper will demonstrate that translation alone is insufficient and that back-translation is an inherently flawed methodology. METHODS: Literature review. RESULTS: Back-translation fails to take account of the complexity of nuances and meanings inherent in language - while we might get back to where we started, it does not imply that we have traveled by the appropriate route. Back-translation can only succeed where straightforward semantic and conceptual equivalents exist in the target language. In such a case, the method is likely to have been an unnecessary expense. A further problem is the reliance on bilinguals. Their social and health status are likely to be higher than average, making them unrepresentative of the patient population. Thus, their use should be restricted to producing a first draft translation. This translation then needs to be assessed by people more typical of the average patient participant, working in their own language. The acceptability of this final version also needs to be tested with relevant patients to confirm comprehension and acceptability. However, adaptation involves both the translation and psychometric testing of an instrument for a new culture. It remains necessary to test the psychometric properties of the translation formally before it is used in a trial. The new version should be shown to have acceptable reproducibility and construct validity. Item response theory should also be applied to ensure that cultural response bias does not exist. CONCLUSIONS: Back-translation by bilinguals cannot guarantee that a translated questionnaire will be acceptable to the target audience. Translation alone is not sufficient to ensure that the new language version is suitable for use in a clinical trial.