The growing complexity of equipment and systems has motivated the search for automated methods of fault diagnosis. Fault diagnosis represents the process of identifying the origin of a fault through the observation of a series of effects that it causes in the system. The method proposed in this paper for system fault diagnosis takes advantage of two very different techniques: Bayesian networks (BN) and systems modeling language (SysML). SysML allows the modeling of requirements, structure, behavior and parameters to provide a robust description of a system, its components, and its environment. This system model is used, in the proposed method, to obtain the BN graph in a novel structured procedure. The BN graph obtained must, in turn, present the components that are most likely responsible for a certain fault of the system under study. The BN model uses components reliabilities to solve the diagnosis problem. A case study of a water storage system is presented and it shows how the method can contribute to an assessment of the monitoring process of a system even in the early stages of its design. With this kind of information, the designer can assess the need for changes in the system to make it more reliable or better monitored.