In this paper, graph of a process (GP) is introduced as a new formalization of causal graph useful in fault diagnosis. Faults are directly incorporated into the model. In this paper, we propose the concept of model structure (MS), which can be used for building fuzzy and neural models for fault detection. Algorithms for finding all MSs and methods for determining faults-symptoms relation are developed. This method is applicable to the design of model-based diagnostic systems, when only basic knowledge of the process to be diagnosed is available. No mathematical model is needed. This paper shows that GP can be constructed on the basis of process diagrams and expert knowledge. Models for fault detection can be built automatically from process measurements. Ideas are explained on a three-tank system example. In the last section, our proposed method is compared in detail with other existing approaches using qualitative models.