The identification of defects supporting Partial Discharge (PD) phenomena in insulation systems of HV machines and components, is a fundamental requirement of any diagnostic procedure because different defect typologies can affect in different ways the insulation reliability. Once a defect is correctly identified, it can be monitored and a repair action can be planned to prevent catastrophic failures. An effective identification requires specific definitions, algorithms and validation procedures. An identification method based on the organization of robust markers in a tree-structure, is presented in this paper. Some examples are discussed to show its advantages/limitations in the perspectives of a possible future development of fully automated diagnostic procedures and smart instruments.