Though artificial neural networks have been successfully applied for diagnostic problems of different natures, difficulties arise by applying this approach, especially the most frequently used back propagation (BP) procedure, for large scale technical diagnostic problems. Therefore, some acceleration techniques of the BP procedure are described and investigated in the paper. Some network models isomorphic to conventional pattern recognition techniques are also enumerated, with special emphasis on the Condensed Nearest Neighbour Network (CNNN), which is a new, self-organizing network model with supervised learning ability. The surveyed techniques are analyzed and compared on a diagnostic problem with more than 300 pattern features.