The definition and classification of early clinically apparent osteoarthritis both in clinical situations and in epidemiological surveys remains a problem. Few data exist on the between-observer reproducibility of simple clinical methods of detecting hand and knee osteoarthritis in the population and their sensitivity and specificity as compared with radiography. Two observers first studied the reproducibility of a number of clinical signs in 41 middle aged women. Good rates of agreement were found for most of the clinical signs tested (kappa = 0.54-1.0). The more reproducible signs were then tested on a population of 541 women, aged 45-65, drawn from general practice, screening centres, and patients previously attending hospital for non-rheumatic problems. The major clinical signs used had a high specificity (87-99%) and lower sensitivity (20-49%) when compared with radiographs graded on the Kellgren and Lawrence scale (2+ = positive). When analysis was restricted to symptomatic radiographic osteoarthritis, levels of sensitivity were increased and specificity was lowered. These data show that certain physical signs of osteoarthritis are reproducible and may be used to identify clinical disease. They are not a substitute for radiographs, however, if radiographic change is regarded as the 'gold standard' of diagnosis. As the clinical signs tested seemed specific for osteoarthritis they may be of value in screening populations for clinical disease.