For the description of pain and for the assessment of treatment pain questionnaires are widely employed but few attempts have been made to study systematically the reliability, validity and clinical usefulness of this tool. In the present study a pain questionnaire composed of questions commonly in use was applied to a group of chronic pain patients, and the results were analysed in order to assess the relevance of each individual item for (a) classification of pain, (b) evaluation of treatment, and (c) prediction of outcome. It was found that the usefulness of the items differed markedly in these three respects. Several of the questions yielded classificatory and predictive information whereas only a few seemed to be suitable for evaluation of outcome of treatment. These results indicate that the selection of items must be made with regard to the special purpose for which the pain questionnaire is intended. The present study may serve as a model for how to proceed in designing a pain questionnaire.