To describe the association between late-life body mass index (BMI) and dementia development with a time perspective and to investigate the effect of weight changes on dementia incidence. Three-, 6-, and 9-year follow-up study. The Kungsholmen Project. One thousand two hundred fifty-five subjects aged 75 and older with baseline BMI data available. Cox-regression models were used to estimate hazard ratios (HRs) for dementia detected at different risk periods in relation to baseline BMI. The association between BMI changes and development of dementia after 3 and 6 years was also analyzed. Subjects with a BMI of 25.0 kg/m2 or higher had a lower risk of developing dementia than subjects with a BMI of 20.0 to 24.9 (HR=0.75, 95% confidence interval (CI)=0.59-0.96), even when cases occurring only during the last follow-up period (6-9 years after BMI assessment) were included (HR=0.66. 95% CI=0.40-1.07). Severe BMI loss (>10%) was related to a greater risk of dementia, but this association was present only for dementia cases detected in the subsequent 3 years (HR=2.18, 95% CI=1.27-3.74). This study does not confirm that being overweight in late life is a risk factor for dementia, although a protective effect for a BMI greater than 25.0 is suggested. In addition, BMI loss is confirmed as a marker of incipient dementia. The findings suggest that, from a clinical perspective, the cognitive profile of elderly persons with unexplained weight loss should be considered and that being moderately overweight at older ages might be indicative of good health status.