Twenty-four patients with rheumatoid arthritis were tested in a randomized, double-blind. Latin-square comparison of 250, 750 and 1500 mg of naproxen daily. Each received each dose for 2 wk and baseline disease activity was established during withdrawal of medication before and after the study. Nine standard measures of efficacy were tested at each evaluation. No order effect or change in baseline was found. Total and unbound naproxen concentrations were measured by high-pressure liquid chromatography and equilibrium dialysis, respectively. A linear dose-response relationship (P less than 0.05) was demonstrated between naproxen and joint count, patient's pain assessment, activities of daily living index, physician's global assessment, and grip strength. The relationship to patients' global assessment was of uncertain significance (P less than 0.07). A positive dose to serum level correlation (1, 2, and 12 hr after dose) was apparent (r greater than 0.78). When patients were defined as responders or nonresponders by a summed efficacy score, there was a serum concentration-response relationship; the percentage of responding patients increased with each serum level quartile: 25%, 31%, 59%, and 75%. Patients with a trough total serum naproxen concentration under 18 micrograms/ml did not respond, while 76% of patients with trough total serum concentrations above 50 micrograms/ml responded. No serum naproxen toxicity level relationship was established.