The process of effective diagnosis of the technical state of an object is an essential task related to its utilization. For this purpose, appropriately developed diagnostic systems are used. However, the process of their elaboration is very difficult because it is necessary to consider very wide domain knowledge during this process—for example, information about the object's structure, possible faults, results of risk analysis, and available diagnostic methods. Generally, in the literature, there are no adequate methods for the purpose of designing diagnostic systems. In this article, a new design approach for this purpose is proposed. It is based on some elements of requirement engineering and the selected methods of artificial intelligence. The main assumptions of the proposed design approach rely on the use of sets of requirements to describe the expected functionalities of the diagnostic system. Each of the requirements is connected with proper faults that may occur during the operation of the considered object, is described by a set of attributes, and implies the partial solution of the diagnostic system. A gathered set of requirements is evaluated using the expert system, such as the inference process in Bayesian networks. The structure of these networks depends on the obtained notation of the requirements. As a result of an evaluation of the set of requirements, based on the data that describe the considered object and the assumed criteria that describe the limitations, a subset of requirements that describe the optimal solution of the diagnostic system is isolated.