Although a large number of publications has Appeared, the problem of seizure forecasting in epilepsy remains unsolved up to now. The main problem is the derivation of EEG signal features, which show distinct changes of an impending epileptic seizure, occuring as fieixure precursors. In previous publications [12], [13] several approaches to the feature extraction problem based on Volterra-Syslenis (VS) and Cellular Nonlinear Networks (CNN) [31 have been proposed. Because distinct changes in the behaviour of discrete-time CNN predictor systems have been observed before the onsets of seizures, a deeper analysis of long lasting recordings of EEG signals will be given throughout this contribution.