In a prospective study of 230 episodes of low-back pain presenting in primary care, the natural history of the symptom of low-back pain has been described. Clinical features predictive of outcome have been identified in order to define groups of patients who were relatively homogeneous with respect to the outcome of the episode. A Disability Questionnaire performed more satisfactorily as an outcome measure than either absence from work or a simple pain-rating scale. Guidelines for future trials of treatment of back pain in primary care are described.