Nonhierarchical cluster analysis was used to classify 92 patients with rheumatoid arthritis drawn from a community rheumatology practice into 5 groups on the basis of biochemical measures and disease indices. The major differentiating variables were the number of active joints, number of damaged joints, overall disease activity, extraarticular complications, and history of joint surgery. Although the 5 subclasses were equivalent on measures of psychological functioning, they differed systematically on such health outcome measures as mobility, physical activity, and dexterity. Relationships between the taxonomy produced through cluster analysis and conventional classifications are discussed, and directions for further investigation are noted.