Performance of rheumatoid arthritis (RA) classification by the 2010 American College of Rheumatology (ACR)/European League against Rheumatism (EULAR) criteria, compared to the 1987 ACR criteria, has not been assessed in population-based cohorts in which disease identification is by mailed questionnaire. Women followed in the Nurses' Health Study and Nurses' Health Study II cohorts self-reported new doctor-diagnosed RA on biennial questionnaires. Two RA experts reviewed medical records of 128 new RA self-reports to obtain individual 1987 and 2010 criteria and arrived at a consensus opinion. We compared agreement in classification by the two criteria sets (kappa), and calculated sensitivity and specificity, with reviewers' opinion as gold standard. Ninety-eight (77%) participants were classified as RA by reviewers' consensus opinion; 98 (77%) fulfilled 1987 criteria, while 79 (63%) fulfilled 2010 criteria. Seventy-two (56%) were classified as RA by both sets, 21 (16%) by neither, 26 (20%) by only 1987 criteria, and 9 (7%) by only 2010 criteria. Kappa for concordance was 0.36 (95% CI 0.20-0.53). Compared to reviewer's opinion, sensitivity and specificity were 0.93 and 0.77 for 1987 criteria, and 0.79 and 0.87 for 2010 criteria. Participants fulfilling 1987 criteria only were more likely to be seronegative. In these prospective population-based cohorts, significant discordance between 1987 ACR and 2010 ACR/EULAR criteria for classifying RA exists. Using the 2010 ACR/EULAR criteria alone had decreased sensitivity, and seronegative RA cases would be excluded in particular. Combined use of both will be necessary to maximize inclusion and allow sensitivity analyses.