A real-time diagnosis system for the detector used in the heavy ion superconducting spectrometer (HISS) experiments is examined. The system is multileveled, combining a single monitoring level based on statistical methods with two model-based diagnostic levels, one operating on structural information and the other using both structural and behavioral models. The model-based approach mitigates the combinatorial complexity of rule-based systems, and the multilevel nature of the architecture mimics the way an expert would diagnose a large-scale system; it takes a broad, shallow look at all the system data and components, produces a limited set of subject components, and then looks at that limited set plus a reduced set of data to identify the most probable suspects.