We systematically reviewed randomized controlled trials (RCTs) of pharmacological therapy in knee osteoarthritis (OA), published between 1966 and August 1994. RCTs were identified by MEDLINE, supplemented by a manual search of reference lists. Qualitative assessment of RCTs was performed using Gotzsche's method; design and analysis features were rated on a scale of 0 (worst) to 8 (best). Heller et al's method was used to compare efficacy of nonsteroidal antiinflammatory drugs (NSAIDs) in comparative trials. A total of 80 RCTs were analyzed (45 involved NSAIDs, 3 analgesics, 5 intraarticular [IA] steroids, 9 biological agents, including IA hyaluronic acid, and 18 mixed modalities, including topical capsaicin). The median design and analysis scores for all 80 RCTs were 2 and 5, respectively. NSAIDs were superior to placebo in all short-term trials, but in the 32 comparative NSAID trials, only five (16%) found significant differences in efficacy. Heller et al's method identified differences in 14 NSAID comparisons; etodolac (600 mg/day) was superior in five of its nine comparisons. Indomethacin and aspirin were the most toxic NSAIDs. IA steroids were superior to placebo in short-term efficacy (< 1 month). Biological agents were superior to placebo and generally well tolerated over a mean follow-up of 48 weeks. Acetaminophen was superior to placebo and was comparably efficacious to low-dose naproxen and ibuprofen (< 2,400 mg/day). The data support the use of acetaminophen, topical capsaicin, IA steroids, IA hyaluronic acid, and NSAIDs in the treatment of patients with knee OA.