A new method for diagnostic knowledge acquisition based on the generalized failure mechanism knowledge (GFMK) has been proposed in order to solve the problem of incompleteness of the diagnostic knowledgebase. The present study consists of two main issues; derivation of the GFMK from actually experienced events and the failure mode estimation based on the plant model. Failure events and objective plant are described by multi-level state transition model and multi-level, multi-attribute model, respectively. A systematic knowledge representation has been obtained through combined use of these two methods. The validity of the failure events representation method and the capability of the GFMK extraction from failure events have been demonstrated by applying proposed methodology to the actually experienced failure events. The failure mode estimation capability has been demonstrated by applying the GFMK to the primary loop recirculation system of boiling water reactor represented by the proposed model.