The aims of the present study were to investigate the association between overweight and cognitive ability and dementia, and to evaluate the usefulness of self-reported body mass index (BMI) in late life and various data sources commonly used in epidemiological studies to identify persons with dementia. Data were drawn from three population-based studies: the Swedish Adoption/Twin Study of Aging (SATSA), Aging in Women and Men: A Longitudinal Study of Gender Differences in Health Behaviour and Health among Elderly (the Gender Study), and the Finnish Lieto Study. In Study I, the agreement between self-reported and measured BMI over time was evaluated among 774 men and women, ages 40 to 88 years at baseline (mean age 63.9) participating in both the questionnaire phase and in-person testing of SATSA. Latent growth curve (LGC) modeling showed a small but significant increase between self-reported and measured BMI (0.02 kg/m²/y) over time, which would probably not affect the results if self-reported BMI were used as a continuous variable in longitudinal research. In Study II, the agreement between dementia diagnoses from various sources and dementia diagnoses set at a consensus conference was evaluated. Among the 498 elderly people ages 70 to 81 at baseline (mean age 74.5) enrolled in the Gender Study, 87 were diagnosed with dementia during an eight-year period. Review of medical records and nurse evaluations yielded the highest sensitivity (0.83 and 0.80, respectively) and a high specificity (0.98 and 0.96), indicating that these sources might be good proxies of dementia, while data extraction from the Swedish Inpatient Discharge Registry underestimated the prevalence of dementia (sensitivity 0.26). In Study III, the association between being overweight in midlife and cognitive ability in late life was examined in SATSA. The 781 participants ages 25 to 63 at baseline (mean age 41.6) in 1963 or 1973 self-reported their height and weight. From 1986 until 2002, they were assessed five times using a cognitive test battery. LGC models showed that people with higher midlife BMI scores had significantly lower cognitive ability and a significantly steeper decline than their thinner counterparts, an association that persisted when those who developed dementia during the study period were excluded from the analysis. This finding indicates that being overweight might affect cognitive ability independently of dementia. In Study IV, the association between BMI and dementia risk in older persons was described among 605 persons without dementia and ages 65 to 92 at baseline (mean age 70.8) in the Lieto Study. Among these, 86 persons were diagnosed with dementia during eight years of follow-up. Cox regression analyses indicated that for each unit increase in BMI score, the risk of dementia decreased 8% (hazard ratio = 0.92, 95% confidence interval = 0.87–0.97) and the association remained significant when individuals who developed dementia during the first four years of follow-up were excluded from the analyses. This result suggests that low BMI scores are present almost a decade before clinical dementia onset.