The 1987 American College of Rheumatology (ACR) criteria for the classification of rheumatoid arthritis (RA) were clinically assessed. These criteria do not include findings of synovial fluid (SF) analysis and require no exclusion criteria. We have studied sequential patients with arthritis seen in four rheumatology centers in the Philadelphia area. Classifications by the ACR criteria were compared with our clinical diagnoses. Two hundred ninety eight patients were evaluated, 113 with RA and 185 with other diagnoses. Classifications as RA by the ACR criteria corresponded to our clinical diagnosis in 95% of the cases, corroborating the high sensitivity previously reported. However, we found a lower specificity (73%) than that reported (89%). False positive classifications as RA were found in 71% of patients with psoriatic arthritis, 48% of patients with SLE, and 31% of patients with gout. The specificity could be improved to 89% by excluding disorders with obvious distinguishing extraarticular features such as psoriasis or by SF findings of monosodium urate crystals. Awareness of these possible sources of confusion will further increase the teaching and epidemiologic value of these useful simplified criteria.