Introduction: An ACR/EULAR task force released new criteria to classify rheumatoid arthritis at an early stage. This study evaluates the diagnostic performance of these criteria and algorithms by van der Helm and Visser in REACH. Methods: Patients with symptoms ≤12 months from REACH were used. Algorithms were tested on discrimination, calibration and diagnostic accuracy of proposed cut-points. Two patient sets were defi ned to test robustness; undifferentiated arthritis (UA) (n=231) and all patients including those without synovitis (n=513). The outcomes evaluated were methotrexate use and persistent disease at 12 months. Results: In UA patients all algorithms had good areas under the curve 0.79, 95% CI 0.73 to 0.83 for the ACR/EULAR criteria, 0.80, 95% CI 0.74 to 0.87 for van der Helm and 0.83, 95% CI 0.77 to 0.88 for Visser. All calibrated well. Sensitivity and specifi city were 0.74 and 0.66 for the ACR/EULAR criteria, 0.1 and 1.0 for van der Helm and 0.59 and 0.93 for Visser. Similar results were found in all patients indicating robustness. Conclusion: The ACR/EULAR 2010 criteria showed good diagnostic properties in an early arthritis cohort refl ecting daily practice, as did the van der Helm and Visser algorithms. All were robust. To promote uniformity and comparability the ACR/EULAR 2010 criteria should be used in future diagnostic studies. Copyright Article author (or their employer) 2011.