Many limitations and conflicting results have cast serious doubts on the validity of cerebrospinal fluid tau and Abeta42 levels for the biological diagnosis of Alzheimer's disease, particularly extreme variations of the reference limits found by unrelated groups as a consequence of different reference populations used. In this study, we addressed the issue of defining reference limits for cerebrospinal fluid tau and Abeta42 in healthy adult individuals. One hundred and five neurologically intact subjects were enrolled according to strict inclusion criteria, 10 of them with autopsy confirmation of brain integrity. All cerebrospinal fluid samples were similarly and optimally processed as were the dosage methods used and the statistical analyses performed. A robust correlation with age was demonstrated for Abeta42 but not for tau. For tau, we found that an upper cut-off value of 443 ng/l allowed 95% of the subjects to be correctly classified as normal. For Abeta42, a lower cut-off value of 90 ng/l allowed a correct classification of 90% of the subjects. However, a large variance of the reference values, partly explained by the potential contamination of the reference population with presymptomatic dementia patients, may limit the use of reference limits based on living subjects. We propose that the issue of defining reference limits for both cerebrospinal fluid tau and Abeta42 may ultimately be settled by studying large numbers of autopsy-proven neurologically intact individuals only.