Background: Physical activity may decrease the risk of dementia; however, previous cohort studies seldom investigated the different types of physical activity and household activities. Our objective was to analyze the links between two physical activity types and dementia in older people. Methods: The study used data from the prospective observational Three-city cohort and included 1550 community-dwelling individuals aged 72 to 87 without dementia at baseline. Physical activity was assessed with the Voorrips questionnaire. Two sub-scores were calculated to assess household/transportation activities and leisure/ sport activities. Restricted cubic spline and proportional hazard Cox models were used to estimate the non-linear exposure-response curve for the dementia risk and the appropriate activity level thresholds. Models were adjusted for possible confounders, including socio-demographic variables, comorbidities, depressive symptoms and APOE genotype. Results: The median age was 80 years, and 63.6% of participants were women. After a median follow-up of 4.6 years, dementia was diagnosed in 117 participants (7.6%). An inverse J-shaped association was found between household/transportation physical activity sub-score and dementia risk, which means that the risk is lowest for the moderately high values and then re-increases slightly for the highest values. The results remained significant when this sub-score was categorized in three classes (low, moderate, and high), with hazard ratios (95% confidence interval) of 0.55 (0.35-0.87) and 0.62 (0.38-1.01) for moderate and high activity levels, respectively. No significant effect was found for leisure/sport activities. Conclusions: The 5-year risk of dementia was significantly and negatively associated with the household/ transportation activity level, but not with the leisure and sport activity sub-score. This highlights the importance of considering all physical activity types in 72 years or older people.