This research is an investigation of model-independent fault diagnosis systems that are intended to be robust, comprehensive, extendible, and practical. The approach is to exploit the underlying probabilistic framework of the system, which makes it possible to incorporate disparate diagnostic algorithms, different sets of data, and a mixture of fault models into a single diagnostic result. The use of heterogeneous fault candidates will allow a diagnosis result that pinpoints problem areas based on data gleaned from the results of logic tests, delay tests, and IDDQ tests. The research results are being verified by testing them using actual industrial IC failure data. Algorithms and software demonstrating the research ideas are being developed.