Recently, prediction models for type 2 diabetes mellitus (T2DM) in older adults (aged a parts per thousand yen55 year) were developed in the KORA S4/F4 study, Augsburg, Germany. We aimed to externally validate the KORA models in a Dutch population. We used data on both older adults (n = 2,050; aged a parts per thousand yen55 year) and total non-diabetic population (n = 6,317; aged 28-75 year) for this validation. We assessed performance of base model (model 1: age, sex, BMI, smoking, parental diabetes and hypertension) and two clinical models: model 1 plus fasting glucose (model 2); and model 2 plus uric acid (model 3). For 7-year risk of T2DM, we calculated C-statistic, Hosmer-Lemeshow chi(2)-statistic, and integrated discrimination improvement (IDI) as measures of discrimination, calibration and reclassification, respectively. After a median follow-up of 7.7 years, 199 (9.7%) and 374 (5.9%) incident cases of T2DM were ascertained in the older and total population, respectively. In the older adults, C-statistic was 0.66 for model 1. This was improved for model 2 and model 3 (C-statistic = 0.81) with significant IDI. In the total population, these respective C-statistics were 0.77, 0.85 and 0.85. All models showed poor calibration (P <0.001). After adjustment for the intercept and slope of each model, we observed good calibration for most models in both older and total populations. We validated the KORA clinical models for prediction of T2DM in an older Dutch population, with discrimination similar to the development cohort. However, the models need to be corrected for intercept and slope to acquire good calibration for application in a different setting.