The diagnosis of Alzheimer's disease (AD) is still made by excluding other disorders with a similar clinical picture. In addition, an analysis of symptoms and signs, blood analyses and brain imaging are the major ingredients of the clinical diagnostic work-up. However, the sensitivity of a clinical diagnosis using these instruments is unsatisfactory and disease markers with high sensitivity and specificity for AD would be a welcome supplement. Ideally, such markers should reflect the pathophysiological mechanisms of AD, that is, according to the currently predominant hypothesis mismetabolism of beta-amyloid and neurofibrillary degeneration. Among several, we have focused on three candidates that have been suggested to fulfil the requirements for biomarkers of AD: beta-amyloid42 (Abeta42), total tau (T-tau) and tau phosphorylated at various epitopes (P-tau). The cerebrospinal fluid (CSF) levels of these proteins reflect the metabolism of these proteins in the central nervous system. Only published articles using established ELISA methods for the quantification of these markers in CSF and preferably also presenting sensitivity and specificity figures have been included in this review. The number of patients included in the different studies varies; some having included only a few patients. Furthermore, diagnostic criteria vary and clinicopathological studies are scarce. However, there are some large studies, including even minor studies, and most have found reduced CSF levels of Abeta42 and increased CSF levels of T-tau in AD. The sensitivity and specificity of these measures are high for separation of AD patients from controls, but their specificity against other dementias is moderate. It increases if P-tau is added. An increasing number of studies suggest that supplementary use of these CSF markers, preferably in combination, adds to the accuracy of an AD diagnosis.