OBJECTIVE : The gait variables that are related to the risk of incident dementia have not yet been clarified. To examine whether quantitative gait variables can predict incident dementia in older adults. METHODS : A prospective study. METHODS : A community setting. METHODS : The study included 4011 participants (54% women; mean age 72.2 years). METHODS : Gait speed, stride length, and stride length variability were measured at baseline. Dementia was prospectively confirmed from monthly updated medical records. RESULTS : The mean follow-up duration was 42.8 months, and there were 245 incident cases of dementia. A receiver operating characteristic curve showed the cut-off value for each gait variable against incident dementia during follow-up using the Youden index (gait speed: 1.12 m/s; stride length: 1.15 m; stride length variability: 3.67%). Based on these cut-off values, Cox proportional hazards regression models showed that after adjusting for covariates, dementia was significantly related to slow gait speed [hazard ratio (HR) 1.49, 95% confidence interval (CI) 1.10-2.01, P = .011], short stride length (HR 1.77, 95% CI 1.29-2.41, P < .001), and high stride length variability (HR 1.58, 95% CI 1.20-2.08, P = .001). Similarly, in a subanalysis stratified by sex, all gait variables were significantly related to the rate of incident dementia during follow-up in both men and women (all P < .05). CONCLUSIONS : Reduced gait function was related to incident dementia, and this association was independent of sex. Quantitative and multifaceted gait variables are valuable for assessing dementia risk.