The aim of this study was to determine the performance of several spectral indices of the EEG (ratios between fast and slow EEG activities) as descriptors of the EEG changes occurring at the onset and during the evolution of Alzheimer's disease (AD). These indices were calculated from quantitative analysis of EEGs recorded in AD patients and from a matched non-demented group of control subjects. One advantage of such indices is to be independent of the absolute value of power spectral densities, which may vary from subject to subject, another being to take into account fast EEG activities. Conventional statistic tests and Receiver Operating Curves (ROC) analysis were performed upon these data to determine the accuracy of the power ratios to discriminate a) between controls and patients (i.e., to detect dementia) and b) between subgroups of patients defined according to the Global Deterioration Scale of Reisberg (GDS). The defined ratios provided a good classification of AD patients for all cerebral regions except the frontal areas, because of eye movement artefacts; the results confirm the increase in slow activities and the concomitant decrease in fast activities early in AD patients. Moreover, our results demonstrate that these indices are adapted tools to perform a good discrimination between demented and non-demented patients in routine clinical practice. We therefore propose the use of these EEG power ratios to discriminate between different stages of Alzheimer's disease, and to perform long-term monitoring of AD patients.