Background. A number of recent epidemiological studies have shown that the prevalence and incidence of dementia are increased in population strata with low compared to high levels of education. This has been explained as a consequence of a greater 'brain reserve capacity' in people with a high level of education. Theoretically, however, brain reserve capacity is better reflected by intelligence than by level of education. Thus, the emergence of dementia will be better predicted by low pre-morbid intelligence than by low education. Methods. This prediction was tested in a population based sample of elderly subjects (N = 2063; age range 65-84; Amsterdam Study of the Elderly) who were followed over 4 years. Dementia was diagnosed using the Geriatric Mental State examination (GMS). Pre-morbid intelligence was measured using the Dutch Adult Reading Test (DART), a short reading test which gives a good estimate of verbal intelligence, and is relatively insensitive to brain dysfunction. The effects of age, gender, occupational level, number of diseases affecting the central nervous system and family history of dementia or extreme forgetfulness were also examined. Results. Logistic regression analysis showed that low DART-IQ predicted incident dementia better than low level of education. A high occupational level (having been in charge of subordinates) had a protective effect. Conclusions. This result supports the brain reserve theory. It also indicates that low pre-morbid intelligence is an important risk factor for cognitive decline and dementia. Use of reading ability tests is to be preferred over years of education as estimator of pre-morbid cognitive level in (epidemiological) dementia research.