An immunodiagnosis model for power transformer faults is put forward in terms of the recognition mechanism between antibody and antigen. The functional model is developed

and the way it diagnoses faults and learns new patterns is analyzed at length. This model is capable of realizing fast diagnosis

multi-resolution identification of transformer faults and new patterns learning. The diagnostic efficiency is proved greatly improved and the disability to learn is thus overcome. Experimental results show that this model is effective and better practicality can be achieved in on-line monitoring.