Background: Extraction of physiological rhythms from Electroencephalography(EEG) data and further their automated analysis are extensivelystudied in clinical monitoring, particularly, to find traces ofinterictal/ictal states of Epilepsy. Methods: Since each rhythm differently influences on brain neuronal activityfor distinguishing between normal and interictal/ictal events, ourproposed methodology measures contribution of therhythms in termsof their stochastic variability extracted from the Short TimeFourier Transform to highlight the nonstationary behavior of the EEGdata. Then, we carry out the variability-basedrelevance analysisby handling the multivariate short--time rhythm representationwithin asubspace framework that searches for a projection maximallybearing input information when preserving only those data thatcontribute most to the brain activity classification. For purposes of neural activity monitoring, we also develop a new relevancerhythm diagram that is aqualitative meaningful evaluation of rhythmvariability throughout long time periods aiming at distinguishingevents with different brain neuronal activity. Results: Evaluation is carried out over two EEG datasets, one of which isrecorded under real noisyenvironment. The method is evaluated forthree different classification problems, each of them addressing adifferent interpretation of the medical problem. We perform a blinded study of 40 patients using the support--vector machineclassifier cross--validation scheme. Obtained results show that theuse of the developed relevance analysis aims at differentiatingnormal, ictal and interictal activities with high accuracy. Conclusions: The proposed approach provides reliable identification of traces ofinterictal/ictal states of Epilepsy. Also, the introduced relevancerhythm diagram of physiological rhythms gives effective support toEpileptic seizure monitoring with the added benefit of easy implementationand clinical interpretation. Developed variability--based