The accurate diagnosis of dementia is difficult because there are no specific physiologic tests. Computer-analyzed electroencephalography (EEG) has shown promise as a tool for the differential diagnosis of dementia, but topographic methods for data collection and analysis have seldom been used. We used these methods to investigate EEG differences among three groups of elderly subjects: those with mild to moderate dementia of the Alzheimer's type (DAT), those with mild to moderate multi-infarct dementia (MID), and age- and sex-matched normal controls. The two groups of demented subjects were distinguished from the control group by a new criterion, the ratio of high-frequency to low-frequency electrical activity in the left temporal region, which was greatly diminished in demented subjects. Using this single variable, all control subjects and 15 of 18 demented subjects were correctly classified (sensitivity 83%, specificity 100%). The three misclassified subjects were among the least impaired subjects with DAT. Examination of coherence, or synchronization of the EEG signal, enabled us to distinguish subjects with DAT from those with MID. Using discriminant analysis of both EEG frequency and coherence, 92% (22/24) of subjects were accurately classified. These results justify further evaluation of topographic, computer-based EEG analysis as a test for the differential diagnosis of dementia.