The complexity of modern industrial processes and the large amount of data available to their operators make it difficult to monitor their status and diagnose potential failures. Although there have been many attempts to apply knowledge-based technologies to this problem, there have not been any convincing success. This paper describes recent experiences with a technology that combines artificial intelligence and simulation techniques for building real-time monitoring and diagnosis systems. A prototype system for monitoring and diagnosing the feedwater system of a nuclear power plant built using this technology is described. The paper then describes several interesting classes of failures that the prototype is capable of diagnosing. (author). 19 refs, 6 figs.