The detection of patterns embedded within a complex, nonstationary, and noisy background activity is a crucial and important task in EEG analysis. The authors present a biologically inspired, analog approach to EEG analysis that is conceptually different from a variety of statistical approaches currently used. A nonlinear, excitable, spatially extended medium that is composed of diffusively coupled model neurons is considered. When EEG recordings are applied as local perturbations to such an excitable neural tissue, the induced transient changes in the dynamics of the perturbed system can be regarded as an instantaneous characterization of transient processes in the brain reflected by the EEG, e.g., in the form of a sequence of correlated dynamical events (patterns). Nonlinear excitable media can be implemented in form of an array of locally coupled integrated analog nonlinear electrical circuits called cellular neural networks, which represent a next evolutionary step in the parallel analog computer architecture. Using cellular neural networks, the authors show that the concept of signal-induced pattern generation allows an almost instantaneous and unsupervised detection of seizure onsets in EEG recordings. In addition, they show that a cellular neural network can be trained in a supervised way to approximate the degree of synchronization in EEG recordings. The resulting pattern-recognition device may be suitable for the prediction of epileptic seizures.