Five hundred patients with established, symptomatic limb joint OA have been recruited into an on-going prospective study of the natural history of the condition. Four hundred and fifteen patients (mean age 65.6 years, female to male ratio 2.05:1) were available for a full clinical and radiographic review 3 years after entry (mean entry to follow-up interval 37.6 months, range 31-41). The majority reported an overall worsening of their condition, although pain severity did not change. There was an overall increase in disability (Steinbrocker) and the use of walking aids in the group but 57 patients (13.7%) improved, 38 of whom had undergone joint surgery. There was a strong correlation between changes in different clinical outcome measures, but none of the baseline variables predicted change over 3 years with the exception of an association between pain severity and subsequent surgery. One hundred and ninety-three of the 415 patients had knee joint disease at entry. One hundred and forty-five of these patients had knee radiographs and full clinical data available from both time points. Some change was seen in 85 of 276 evaluable tibiofemoral joints (30.1%), but only 10 patellofemoral joints. There was a strong correlation between changes in joint space, osteophyte and subchondral bone scelerosis. However, there was no correlation between radiographic and clinical changes. It is concluded that radiographic change may not be a good surrogate for clinical outcome in established OA. This has implications for the design of long-term studies of possible structure modifying agents in OA.