This paper reports on the application of model-based diagnosis techniques to diesel engine management systems within the Brite-EuRam project “Vehicle Model Based Diagnosis”. We discuss some major requirements that have been identified in this application. In particular, it is essential to solve the inherent variant problem, to reason across different physical domains and to fulfill real-time needs for on-board diagnosis. The main foundation of our approach is to use qualitative models, especially qualitative deviation models, which serve as a coherent modeling paradigm for the different domains. In the project, this technology has been implemented and evaluated for on-board diagnosis on two demonstrator vehicles. The paper also discusses further perspectives of the technology for tools supporting the development and implementation of on-board diagnosis.