Patients with rheumatoid arthritis, in stable treatment with methotrexate, penicillamine, or sulfasalazine, were randomized in a double-blind fashion either to continuation of their usual treatment or to placebo. 112 patients were included; 52 patients who refused participation had no more severe disease than the others. The patients felt worse on placebo than on active drug (p = 0.002). The mean differences in number of tender, painful and swollen joints after one month were 2.4 (p = 0.08), 3.0 (p = 0.12) and 2.2 (p = 0.03), respectively. Treatment failure occurred for 42 patients of whom 33 received placebo (p = 0.000,001). There was no difference in the severity of side effects (p = 0.91). The patients guessed their treatment correctly more often than expected (p = 0.02) because of the perceived effect. None of the two observers guessed better than chance, and there were no differences between the observers' evaluations of the joints. The effect of slow-acting antirheumatic drugs was unequivocal and no observer bias occurred.