To establish diagnostic performance of the cerebrospinal fluid (CSF) biomarkers beta-amyloid peptide (Abeta(1-42)), total tau-protein (T-tau) and tau phosphorylated at threonine 181 (P-tau(181P)) compared to clinical diagnosis, biomarker levels were determined in CSF samples from 100 autopsy-confirmed dementia and 100 control subjects. As the control and dementia groups were not age-matched and given the significant associations of biomarker concentrations with age in controls, age-corrected biomarker concentrations were calculated. New models were constructed by means of logistic regression. Using all biomarkers, dementia could be discriminated from controls (sensitivity (S)=86%, specificity (Sp)=89%). T-tau and Abeta(1-42) optimally discriminated Alzheimer's disease (AD) from other dementias (NONAD) and controls (S=90%, Sp=89%). AD was optimally discriminated from NONAD using P-tau(181P) and Abeta(1-42) (S=80%, Sp=93%). Diagnostic accuracy of the latter model (82.7%) was comparable to clinical diagnostic accuracy (81.6%) that was based on a whole clinical work-up (including imaging). Using this model, in cases with clinically doubtful diagnoses, a correct diagnosis would have been established in 4/6 autopsy-confirmed AD and 3/3 autopsy-confirmed NONAD cases. The value of biomarkers in differential dementia diagnosis was shown, using pathological diagnosis as a reference. New models have been developed, achieving sensitivity, specificity and diagnostic accuracy levels, consistently exceeding 80%.