Objective: To evaluate the classification criteria of early rheumatoid arthritis (ERA) and compare the sensitivity and specificity with the criteria of 1987 American College of Rheumatology (ACR) criteria and 2010 ACR/European League Against Rheumatism (EULAR). Methods: Patients from 4 hospitals, aged more than 16 years, with arthritis, whose disease duration was ≤1 year, and with ≥1 joint pain and swelling were enrolled in the study. The indicators including clinical manifestations, laboratory tests and imaging examinations were observed. The ERA patients were dignosed by two experienced rheumatologists based on the clinical features, drug therapy information and radiography features. Results: (1) A total of 325 patients with arthritis were enrolled, including 98 males (30.15%) and 227 females (69.85%), The average age was (47.53±14.44) years, and the median disease duration was 5 (2, 8) months. Finally, 236 patients were dignosed with ERA, and 89 patients were dignosed with other diseases (Non-ERA, including osteoarthritis, reactive arthritis, undifferentiated arthritis, spondyloarthritis, etc). (2) The sensitivity of ERA criteria was 87.29%, and the specificity was 84.37%. The sensitivity was higher than that of 1987 ACR criteria (χ ²=43.641, P<0.001), and had no significant difference compared with 2010 ACR/EULAR criteria (χ²=0.446, P=0.593). But the specificity of ERA criteria was lower than that of 1987 ACR criteria (χ²=4.891, P=0.027), which was not statistically significant compared with 2010 ACR/EULAR criteria (χ²=0.044, P=1.000). (3) In the patients with arthritis whose disease duration was ≤3 months and ≤6 months, the sensitivity of ERA criteria was 81.71% and 86.79%, respectively, both were higher than the 1987 ACR criteria (χ²=7.131, P=0.008; χ²=22.015, P<0.001) and had no statistically difference compared with the 2010 ACR/EULAR criteria (χ²=0.220,P=0.755; χ²=0.473,P=0.491). The differences of the three criteria in specificity were not statistically significant. (4) The three different classification criteria were consistent with the clinical diagnosis, among which the ERA criteria and 2010 ACR/EULAR criteria were slightly higher (Kappa>0.6). The results of the consistency comparison between the three criteria showed that the ERA criteria and 2010 ACR/EULAR criteria had a better consistency (Kappa=0.836). Conclusion: The sensitivity of ERA classification criteria in the diagnosis of ERA was higher than that of 1987 ACR criteria, and was equivalent to that of 2010 ACR/EULAR criteria. There is no significant difference in specificity between these three criteria. The ERA criteria can also identify patients with RA at a very early stage in arthritis with disease duration ≤3 months.