The objective was to assess the performance of the SF-36 health survey (SF-36) in a sample of patients with rheumatoid arthritis (RA) stratified by functional class. The eight SF-36 subscales and the two summary scales (the physical and mental component scales) were assessed for test retest reliability, construct validity and responsiveness to self-reported change in health. In 233 patients with RA, the SF-36 scales were: reliable (intra-class correlation coefficients 0.76-0.93); correlated with American College of Rheumatology (ACR) core disease activity measures [Spearman r = -0.12 (erythrocyte sedimentation rate) to -0.89 (Modified Health Assessment Questionnaire)]; and responsive to improvements in health (standardized response means 0.27-0.9). The distribution of scores on four of the eight subscales (physical function, role limitations physical, role limitations emotional and social function) was clearly non-Gaussian. Very marked floor effects were noted with the physical function scale, and both ceiling and floor effects with the other three subscales. The two SF-36 physical and mental component summary scales are reliable, valid and responsive measures of health status in patients with RA. Six of the eight subscales meet standards required for comparing groups of patients, and the physical function and general health scales may be suitable for monitoring individuals. The two scales measuring role limitations have poor measurement characteristics. The SF-36 pain and physical function scales may be suitable for use as patient self-assessed measures of pain and physical function within the ACR core disease activity set.