Background and purpose No study has established a prediction dementia model in the Asian populations. This study aimed to develop a prediction model for dementia in Chinese type 2 diabetes patients. Methods The retrospective cohort study included 27 540 Chinese type 2 diabetes patients (aged 50–94 years) enrolled in the Taiwan National Diabetes Care Management Program. Participants were randomly allocated into derivation and validation sets at a 2:1 ratio. Cox proportional hazards regression models were used to identify risk factors for dementia in the derivation set. Steps proposed by the Framingham Heart Study were used to establish a prediction model with a scoring system. Results The average follow‐up was 8.09 years, with a total of 853 incident dementia cases in the derivation set. The dementia risk score summed up the individual scores (from 0 to 20). The areas under the curve of 3‐, 5‐ and 10‐year dementia risks were 0.82, 0.79 and 0.76 in the derivation set and 0.84, 0.80 and 0.75 in the validation set, respectively. Conclusions The proposed score system is the first dementia risk prediction model for Chinese type 2 diabetes patients in Taiwan.