Eighty-nine patients with established OA of the knee joint, already on regular NSAIDs for joint pain, were randomly allocated to receive 100 mg/day of slow release diclofenac (45 patients) or matching placebo (44), in place of their NSAID, for 2 years. Thirty-eight patients withdrew or dropped out of the study. The major causes for withdrawal were lack of efficacy (three active, 12 placebo, P < 0.01) or side effects (six active, five placebo), and most withdrawals occurred within the first 6 months. Long term follow up of these patients was not possible. Fifty-one patients completed the study (31 active, 20 placebo), 35 of whom reported that they were the same or better at the end of the 2-year period than at the beginning. Most of the recorded clinical parameters showed little or no change over 2 years in these 51 subjects, and in 70% there was no detectable change in the radiographs. We conclude that long term placebo-controlled trials are both feasible and ethical in knee OA, but that conventional clinical and radiographic techniques detect very little change in joint structure or function over a 2-year time period. This may reflect the insensitivity of the methods used to assess progression rather than absence of change. The fact that 20 of 44 patients changed from an NSAID to placebo completed the 2-year study without any symptomatic penalty indicates that not all patients entered needed or responded to NSAIDs.