A multi-grade fuzzy evaluation method for power transformer using fuzzy theory is proposed. Firstly， a layered evaluation index system on the base of preventive experiments， including working surroundings， operation history， maintenance records，accessories section and so on is built. Then， the relative impairment degree is introduced to describe the relative grade of transformers changing from actual conditions to faults. Taking the relative impairment grade as inputs， a neural network is applied to compute the member function of dissolved gases. The member functions for qualitative indices and the quantitative indices are respectively determined with helps of a fuzzy statistic method and a fuzzy distribution method. An operation condition evaluation model for power transformer， which evaluates the state of every grade and gives a comprehensive assessment of transformer， is established on the base of the fuzzy synthetic judgment. Example analysis shows the condition assessment model proposed is sound and effective.