BACKGROUND: In 2010, new classification criteria for rheumatoid arthritis (RA) were developed. OBJECTIVE: To assess agreement between 1987 American College of Rheumatology (ACR) and 2010 ACR/European League Against Rheumatism (EULAR) criteria and the potential source of discordance, based on ESPOIR cohort data. METHODS: 813 early arthritis patients were included in ESPOIR between 2002 and 2005. Between-criteria agreement was based on the κ coefficient. Discordance was explored by logistic regression. RESULTS: Data for 811 patients were available, with their main characteristics as follows: women 77%, swollen joint count 7.2, tender joint count 8.4, disease activity score in 28 joints 5.2, rheumatoid factor 46%, anticitrullinated protein antibody (ACPA) 39%, structural damage 22%. At baseline, 579 (71.4%) patients met the 1987 ACR criteria and 641 (79.0%) the 2010 criteria. Agreement at baseline was discordant for 168 patients: 115 satisfied the 2010 criteria and 53 the 1987 criteria. Concordance between the two sets was fair, with a κ coefficient of 0.45 and 0.42 at baseline and year 2, respectively. The main sources of discordance were the number and symmetry of joint involvement, as well as ACPA status. CONCLUSION: 2010 ACR/EULAR criteria identified more patients with RA than did 1987 criteria. The 2010 criteria failed to identify RA patients with symmetrical seronegative arthritis and limited joint involvement.