An intelligent event-oriented diagnosis methodology and diagnostic system architecture is proposed in this paper. The dynamic behaviour of the process system in different faulty modes is described by cause-consequence event sequences represented by coloured Petri nets (CPNs). The dynamic model of the system is assumed to be partially unknown and it is refined using the observed event sequences by a learning method. The real-time diagnosis operates on the CPN model of the system and on the expected operating procedures comparing observed event sequences to the model-based prediction. The diagnostic system architecture consists of an event archivator, a data structure generator, a real-time diagnostic monitor and a learning subsystem.