OBJECTIVE: To compare the diagnostic accuracy of the 2010 American College of Rheumatology (ACR)/European League Against Rheumatism (EULAR) and 1987 ACR criteria for rheumatoid arthritis (RA) in a cohort of patients with recent-onset arthritis followed-up for 10 years. METHODS: One hundred and sixty-four patients with recent-onset arthritis of less than 1 year's duration were included prospectively between 1995 and 1997. The diagnosis of RA was defined as having a diagnosis of RA made by the office-based rheumatologist 10 years after enrolment. We compared the sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of the criteria sets at baseline. RESULTS: At baseline, 60 of the 164 patients had alternative diagnoses better explaining the arthritis and 13 had erosions typical for RA; of the 91 remaining patients, 33 had at least 6 ACR/EULAR points (indicating definite RA), and 58 had fewer than 6 points. The ACR/EULAR criteria had a quite similar sensitivity than the 1987 ACR criteria (33/57 [57.9%] for ACR/EULAR criteria vs 34/57 [59.6%] for the 1987 ACR criteria), but higher specificity, PPV, and NPV (95/107 [88.8%], 34/46 [73.9%], and 95/118 [80.5%], respectively) than the 1987 ACR criteria (80/107 [74.8%], 33/63 [52.4%], and 80/104 [76.9%], respectively). CONCLUSION: ACR/EULAR criteria performed substantially better than ACR 1987 criteria for predicting a diagnosis of RA after 10 years. Much of the improvement was ascribable to the use of exclusion criteria. BULLET POINTS: (1) The ACR/EULAR criteria had the same sensitivity, but higher specificity, PPV, and NPV than the 1987 ACR criteria; (2) Much of the improvement was ascribable to the use of exclusion criteria.