Much of the existing studies on the built environment and type 2 diabetes are cross-sectional and prone to residential self-selection bias. Using multilevel logistic regression analysis of 36,224 participants from a longitudinal study, we examined whether perceived built environment characteristics are associated with type 2 diabetes. We found that the odds of diabetes incidence varied geographically. Those who reported that there were no local amenities and reported day- and night-time crime rates made walking unsafe in the neighbourhood had higher odds of developing incident type 2 diabetes. These associations persisted after accounting for some predictors of residential self-selection. More longitudinal studies are needed to corroborate the findings. Changing the features of the residential built environment may be an important point of intervention for type 2 diabetes prevention.