The paper proposes an approach to functional reasoning for diagnosis, based on an ontology that is aimed at providing a more formal physical foundation to functional knowledge. The approach (called FDef, i.e. functional diagnosis with efforts and flows) is analysed from several perspectives: reasoning strategy, modelling, formalisation of the reasoning activities and of the entities involved, architecture, comparison with related work, main assumptions and limitations. Furthermore, the paper shows in detail how the minimum entropy approach to measurement prescription can be simplified and used in FDef. Several diagnostic examples are considered and discussed in the paper.