Background: Some studies have found transportation noise to be associated with higher diabetes risk. This includes studies based on millions of participants, relying entirely on register-based confounder adjustment, which raises concern about residual lifestyle confounding. We aimed to investigate associations between noise and type 2 diabetes (T2D), including investigation of effects of increasing confounder adjustment for register-data and lifestyle. Methods: In a cohort of 286,151 participants randomly selected across Denmark in 2010–2013 and followed up until 2017, we identified 7574 incident T2D cases. Based on residential address-history for all participants linked with exposure assessment of high spatial resolution, we calculated 10-year time-weighted mean road and railway noise at the most (LdenMax) and least (LdenMin) exposed façades and air pollution (PM_2.5). We used Cox models to calculate hazard ratios (HR) with increasing adjustment for individual- and area-level register-based sociodemographic covariates, self-reported lifestyle and air pollution. Results: We found that a 10 dB increase in 10-year mean road LdenMin was associated with HRs (95% CI) of 1.06 (1.02–1.10) after adjustment for age, sex and year, 1.08 (1.04–1.13) after further adjustment for register-based sociodemographic covariates, 1.07 (1.03–1.12) after further lifestyle adjustment (e.g. smoking, diet and alcohol) and 1.06 (1.02–1.11) after further PM_2.5 adjustment. For road LdenMax, the corresponding HRs were 1.07 (1.04–1.10), 1.05 (1.02–1.08), 1.04 (1.01–1.07) and 1.03 (1.00–1.06). Railway noise was associated with HRs of 1.04 (0.98–1.11) for LdenMax and 1.02 (0.92–1.12) for LdenMin after adjustment for sociodemographic and lifestyle covariates and PM_2.5. Conclusions: Long-term exposure to road traffic noise was associated with T2D, which together with previous literature indicates that T2D should be considered when calculating health impacts of noise. After sociodemographic adjustment, further lifestyle adjustment only changed HRs slightly, suggesting that large register-based studies with adjustment for key sociodemographic covariates can produce reliable results.