Quantitative topographical EEG was compared with regional glucose metabolism measured by PET with respect to the sensitivity in the classification of mild to moderate dementia. In 24 patients with probable Alzheimer's disease (DAT), 19 patients with vascular dementia (VD) and 15 age-matched healthy controls, global and regional EEG and PET data were analyzed. The metabolic ratio between typically affected and non-affected regions differentiated between DAT and VD (P < 0.001) as well as between DAT and normal controls (P < 0.001) even for the subgroup of mild dementia. In contrast to PET, global EEG changes were more sensitive than regional alterations for the classification into the respective groups. Relative theta power was most sensitive for the differentiation of demented patients irrespective of type of normal controls (P < 0.01), whereas OCC/FR alpha ratio (occipital divided by frontal power) separated between dementia types (P < 0.01) as well as between DAT and normals (P < 0.05). Additionally, EEG may help to grade severity especially in DAT. Combined use of EEG and PET was more discriminative and reached higher diagnostic specificity than each test individually. These results suggest that EEG and PET are complementary diagnostic procedures for the differentiation and classification of dementias.