In a multicentre trial 456 selected patients with low back pain were randomly allocated to one of four treatments-manipulation, definitive physiotherapy, corset, or analgesic tablets. Patients were reassessed clinically after three weeks' treatment and again after a further three weeks. Questionnaires were used to find out the patients' condition three months and one year after admission to the trial. There were never any important differences among the four groups of patients. A few patients responded well and quickly to manipulation, but there was no way of identifying such patients in advance. The response to a corset was slow, but the long-tern effects were at least as good as those of the other treatments. Patients treated only with analgesics fared marginally worse than those on the other three treatments. There is no strong reason, however, for recommending manipulation over physiotherapy or corset.