How can diagnosis and prognosis systems be improved in the presence of uncertainty in test results? How can these uncertainties be identified and modeled? Can diagnosis be improved as a result of understanding these uncertainties? These questions represent the core problems to be explored in this paper. Specifically, we explore the process by which instrument uncertainty can be used to determine conditional probabilities of potential diagnoses given test results generated by these instruments. We then use that information to construct "Bayesian belief networks" with specific goal of maximizing diagnostic accuracy while minimizing construction complexity, and computational complexity. We then extend the ideas presented for Bayesian diagnosis to the prognostic, or predictive diagnostic problem