The EEG research technique on the basis of autoregressive models construction and Granger causality estimation by experimental data are described in this article. The EEG is written down from the brain of WAG/Rij rats, which are absence-epilepsy contaminated. The EEG episodes well enough described in terms of small order linear display along with the episodes with expressed nonlinearity are revealed during the analysis. The EEG episodes ordering is spent in accordance with the model parameters received and physiological condition of the animals. It is reveled that dynamics becomes easier to predict in comparison to the condition of passive wakefulness before the epilepsy attack. Deep brain structures influence enhancement on the cortex and brain structures influence decrease on the reticular kernel before the attack is established.