A realistic analysis of the criteria used by rheumatologists in evaluating the progress of patients suffering from rheumatoid arthritis must be based on actual clinical judgements rather than on expressed opinions. A randomly selected 15% sample of British rheumatologists (48) recorded judgements on the progress of 50 'paper' patients, based on data taken from actual patients participating in clinical trials. The rheumatologists differed markedly in their assessments of the progress of disease, with serious disagreements even when only 'clinically important' changes were considered. Some clinicians showed little consistency in their judgements of duplicate cases. Multiple regression analysis of the patient data in relation to the disease assessments provided a model of each clinician's judgement policy. These judgement policy models showed that the differences in clinical assessment were greater than could be explained by the inconsistent application of similar assessment policies, and were a consequence also of differences in the underlying judgement policies themselves. Judgements related more closely to changes in ESR and other process measures than to changes in functional ability.