The automated detection of electrographic seizures in the neonatal EEG is a difficult, unsolved problem because of the variety of seizure patterns and the large number of seizure-like artifacts and non-seizure rhythmic EEG events. In this paper we present an architecture and methodology for such a detection system designed around a combination of signal processing, pattern recognition, heuristic rules, and neural networks. We believe that this hybrid approach offers the best chance for reliable automated detection of neonatal seizures.