A general outline of the development of PD diagnostics is given with emphasis on problems of the validity of underlying models and the interpretation of results, because PD diagnostics is hardly ever direct and instead rests on a chain of inferences. Some major classical approaches are reviewed in this context, with a discussion of certain problems that also are significant for modern approaches. After a discussion of primary PD events and resulting degradation, some basic matters regarding measurement and testing are described and their problems are discussed. Particular attention is then paid t o various approaches to PD pulse spectroscopy and its application to product diagnostics, particularly after algorithmic processing of the recorded data. The paper concludes with a discussion of conclusions that may or may not be drawn validly from PD diagnostics. A number of warnings and some suggestions of further development are given.