Alzheimer's disease (AD) is the most frequent form of dementia in western countries. The rise in life expectancy will likely increase its prevalence, since ageing is the greatest known risk factor. Although an early and accurate identification is critical, low diagnostic accuracy is currently reached. Hence, the aim of the present study was to analyse the spontaneous magnetoencephalographic (MEG) activity from 148 channels in 20 AD patients and 21 healthy controls to extract discriminating spectral features. Relative power (RP) was calculated in conventional frequency bands and several ratios were defined to emphasise the differences in its distribution. Both RP values and spectral ratios were transformed with a principal component analysis to summarise information with minimal loss of variability. AD patients showed a significant increase of RP(delta) and RP(theta), along with a decrease of RP(beta) and RP(gamma). The most significant differences were reached by spectral ratios using the beta band. Specifically, we obtained 75.0% sensitivity, 90.5% specificity and 82.9% accuracy (linear discriminant analysis with a leave-one-out cross-validation procedure), together with a p-value lower than 0.001 (one-way analysis of variance with age as a covariate) using the [RP(alpha)+RP(beta(1))+RP(beta(2))+RP(gamma)]/[RP(delta)+RP(theta)] ratio. The spectral ratios also showed a higher correlation with the severity of dementia than individual relative power measures. Our results suggest that the spectral ratios could be useful descriptors to help in the AD diagnosis, since they effectively summarise the slowing of the AD patients' MEG rhythms in individual indexes and correlate significantly with the severity of dementia.