Jose Faria holds a degree in electrical engineering from the Faculty of Engineering of the University of Porto, Portugal (1983) and a Ph.D. also from FEUP (1996). The Ph.D. focused on reliability modelling, analysis and evaluation of the integrated management information system of a large European car manufacturer. He is currently an Assistant Professor at FEUP and his research interests include reliability and quality analysis and business operation systems. Eusebio Nunes holds a degree in production engineering from the School of Engineering of the PhD from the Faculty of Engineering of the University of Porto, Portugal (2005). His PhD thesis addressed reliability evaluation of non-Markovian system with fuzzy parameters. His research interests include reliability analysis methods and quality management systems.Abstract: This paper introduces the rationale and the fundamental elements and algorithms of a reliability engineering methodology, and discusses its application to the design of a large, multi-cell and heterogeneous production system with just-in-time (JIT) deliveries. The failure analysis and the non-reliability costs assessment of such systems is a complex task. In order to cope with such complexity, a two level hierarchical modelling and evaluation framework was developed. According to this framework, the internal behaviour of each manufacturing cell and the overall flow of materials are described, respectively, by local and global models. Local models are firstly obtained from the failure and repair processes of the manufacturing equipment. Then, these models are combined with the failure propagation delays introduced by the work-in-process buffers in order to obtain the system level model. The second part of the paper addresses several design issues of the production system that directly impact the reliability of the deliveries, such as the layout of the plant, the redundancy of the manufacturing equipment and the capacity of the work-in-process buffers. A distinctive feature of the reliability evaluation algorithm resides on the ability to deal with reliability models containing stochastic processes with generalized distributions. This fundamental requirement comes from the fact that repair and failure propagation processes typically present hyper-exponential distributions, e.g., lognormal distributions, that can't be assessed using the conventional reliability techniques. The paper will also explain how the behavioural and structural characteristics of JIT production systems were explored in order to implement effective evaluation algorithms that fit the requirements of this class of systems.