Forty-six patients with recalcitrant rheumatoid arthritis entered a trial encompassing a 2-week inpatient period plus a 16-week, randomized double blind, parallel study comparing placebo, 5 mg/m2 and 10 mg/m2 oral weekly methotrexate (MTX). An additional 6 patients, given 20 mg/m2 MTX, contributed to the toxicity, but not the efficacy analysis. All patients had "failed" either gold or D-penicillamine. A linear dose response relationship (placebo vs 5 mg/m2 vs 10 mg/m2) was found for 5 of 11 outcome variables: patient pain and patient global scale, physician global scale, joint tenderness count and activity of daily living scale (p less than 0.05 for each). Gastrointestinal toxicity (p = 0.002), dyspepsia (p less than 0.03) and stomatitis (p less than 0.09) occurred more commonly with MTX, and a general trend, although not significant, was found toward a dose toxicity relationship.