This paper presents a novel approach that formulates different types of diagnostic problems similarly. The main idea is the reformulation of the diagnostic procedures as P-graph models. In this way the same paradigm can be applied to model different aspects of a complex problem. The idea is illustrated by solving the probabilistic diagnosis problem in multiprocessor systems and by extending it with some additional properties. Thus, potential link errors and intermittent faults are taken into consideration and the comparator based diagnostics is formulated including potential comparator errors.