In the event of a partial or total blockage of a liquid-metal fast breeder reactor core subassembly, a boiling zone may be created. Acoustic signals from such a zone could provide a means of early detection of accident conditions. A three-step method, based on pattern recognition techniques, is described and used to analyze data from three experiments that simulate core cooling fault conditions. This method is shown to be capable of detecting the abnormal situation in each of the experiments analyzed.