Background. It is still not clear whether a low level of education increases the risk of developing Alzheimer's disease (AD). Two common problems in cohort studies involving an elderly population and a two-step diagnostic procedure are the loss to follow-up without data on the presence of AD, and the fact that, in general, people with higher levels of education perform better on traditional cognitive tests, such as the Mini-Mental State Examination (MMSE). Both phenomena may lead to misclassification, resulting in a biased association between level of education and AD. This study investigated to what extent these selection mechanisms may influence this association. Methods. In the community-based Amsterdam Study of the Elderly (AMSTEL) a cohort at risk for AD was selected of 3778 people aged 65-84 years. Level of education was expressed in two categories: low (primary education or less) versus high (partial secondary education to completed university education). At follow-up, a subsample of elderly people was selected for further diagnostic evaluation, using a memory test in addition to the MMSE. Clinical diagnoses of AD were made according to DSM-IV (Diagnostic and Statistical Manual of Mental Disorders) criteria. To examine the extent to which loss to follow-up may have affected the results, a sensitivity analysis was performed comparing two extreme possibilities. Furthermore, to examine to what extent use of the MMSE only may have affected the results, the observed odds ratio (OR) was compared with the OR based on only those AD patients who were selected for diagnostics with the MMSE alone. Results. After an average of 3.2 years, 77 people had developed AD. Multivariate logistic regression analyses indicated that a low level of education was associated with incident AD (OR adjusted for age and sex 2.09; 95% CI: 1.29-3.38). The results of the sensitivity analysis still indicated that a low level of education was associated with incident AD. Screening with only the MMSE led to a higher OR than the one observed. Conclusion. Selective attrition and use of cognitive screening tests that are associated with educational level may influence the strength of the association between a low level of education and incident AD; however, it appears that these influences cannot completely explain this association.