OBJECTIVES: To determine whether a home based exercise programme can improve outcomes in patients with knee pain. DESIGN: Pragmatic, factorial randomised controlled trial of two years' duration. SETTING: Two general practices in Nottingham. PARTICIPANTS: 786 men and women aged >/=45 years with self reported knee pain. INTERVENTIONS: Participants were randomised to four groups to receive exercise therapy, monthly telephone contact, exercise therapy plus telephone contact, or no intervention. Patients in the no intervention and combined exercise and telephone groups were randomised to receive or not receive a placebo health food tablet. MAIN OUTCOME MEASURES: Primary outcome was self reported score for knee pain on the Western Ontario and McMaster universities (WOMAC) osteoarthritis index at two years. Secondary outcomes included knee specific physical function and stiffness (scored on WOMAC index), general physical function (scored on SF-36 questionnaire), psychological outlook (scored on hospital anxiety and depression scale), and isometric muscle strength. RESULTS: 600 (76.3%) participants completed the study. At 24 months, highly significant reductions in knee pain were apparent for the pooled exercise groups compared with the non-exercise groups (mean difference -0.82, 95% confidence interval -1.3 to -0.3). Similar improvements were observed at 6, 12, and 18 months. Regular telephone contact alone did not reduce pain. The reduction in pain was greater the closer patients adhered to the exercise plan. CONCLUSIONS: A simple home based exercise programme can significantly reduce knee pain. The lack of improvement in patients who received only telephone contact suggests that improvements are not just due to psychosocial effects because of contact with the therapist.