The incidence of rheumatoid arthritis (RA) was compared in 2 cohorts of women aged 35-64. One consisted of 1,075 estrogen replacement therapy (ERT) users and the other was 3,251 women from general practice registers. Screening detected 32 cases; 8 postmenopausal control and 6 ERT women developed RA during the study period 1982-1986. This produced incidence rates of 19.7/10,000 and 12.3/10,000 years of observation for ERT and controls, respectively. The relative risks for ERT was 1.62 (95% CI 0.56-4.74) and reduced towards unity after adjustment for potential confounders. Despite the wide confidence interval, our data do not support the previous observation of a 4-fold reduction in RA incidence in ERT users. Indeed the incidence rate in the exposed group in this study exceeded current population estimates of RA incidence in postmenopausal women. We believe that the high incidence rates could be best explained by the self-selection for estrogen therapy at the menopause of those with undiagnosed joint symptoms. These findings underscore the difficulties in elucidating the relationship between ERT and RA.