Abstract: In order to make vibration diagnosis of incipient failure and to bring about on-line condition monitoring for mechanical structures of machines, equipments and installations or products, it is necessary to choose vibra-. tion characters sensitive to failure as observed signals for diagnosis procedure and some practically convenient properties must be required for any kind of diagnosis method. Based on this consideration, the concept "se-nsitivety and sensitive domain are defined and emphasis is laid on the description of two new diagnosis methods which are pursued through frequency response function （FR function） in sensitive domain, i. e. FR function-difference pattern recognition diagnosis method and physical parameter-change identificaion diagnosis method. The former can be used to judge whether failures happend in structures and to classify the failures into simple types. The latter is capable of determining both the level and position of failure with the aid of the least square filter technique. Only the single point of output data is needed when the diagnosis is made by both methods. It is shown by experiments that the methods are effective.It is also discussed that the dynamic re-sponse spectrum of structures can be employed as a substitute for FR functions when they are not easy to obtain in some on-line monitoring cases,