Objective: To estimate the incidence of type 2 diabetes among adults in China, evaluate the influence of data from different sources on incidence estimation, and analyze the consistency of such differences across population subgroups. Methods: Based on participants in the second (August 2013 to September 2014, n=23 438) and third (August 2020 to December 2021, n=23 326) resurveys of the China Kadoorie Biobank, new-onset type 2 diabetes cases during follow-up were identified separately through three approaches: field survey (combining questionnaire-based self-report and blood glucose testing), questionnaire-based self-report, and follow-up monitoring through linkage to multiple healthcare databases. Age-standardized incidence rates were calculated using the direct standardization method. Generalized linear mixed-effects models were used to estimate incidence rate ratios (IRRs) and 95%CIs comparing incidence ascertained through field survey versus follow-up monitoring. Results: Among participants in the second resurvey, the crude incidence rates (/1 000 person-years) estimated using field survey, questionnaire-based self-report, and follow-up monitoring were 8.4 (95%CI: 7.9-8.9), 4.3 (95%CI: 3.9-4.6), and 2.8 (95%CI: 2.5-3.1), respectively. The corresponding crude incidence rates(/1 000 person-years) in the third resurvey were 8.3 (95%CI: 7.9-8.6), 6.2 (95%CI: 5.9-6.5), and 5.8 (95%CI:5.5-6.1), respectively. The age-standardized incidence rate (/1 000 person-years) was 7.6 (95%CI: 7.1-8.2) in the second resurvey and 7.4 (95%CI:7.0-7.8) in the third resurvey based on field survey results. The IRR comparing field survey with follow-up monitoring was 3.27 (95%CI: 2.90-3.70) in the second resurvey and declined to 1.46 (95%CI: 1.37-1.56) in the third resurvey, with this trend being observed consistently across subgroups. In both resurveys, participants with an education level of middle school or above, those with annual household income <20 000 Yuan, those with central obesity, and those without coronary heart disease or stroke showed relatively higher IRRs, indicating that incidence was more markedly underestimated by follow-up monitoring in these subgroups compared with field survey. Conclusion: This study indicated that in large population-based cohorts, integrating multi-source health care data could substantially improve the identification of new type 2 diabetes cases compared with relying on a single data source.