Traditional transformer health state assessment is mainly focused on evaluation guideline and mathematical model establishment. However, the assessment results may be inaccurate due to human factors and low data utilization. A clustering method for transformer health state based on historical information mining is proposed to solve these problems. First, association analysis method is applied to extract transformer historical information and quantify evaluation indexes with confidence level. Then association weights of evaluation indexes are obtained from principal component analysis to correct index clustering space. Finally, Canopy-kmeans two-layer clustering method is utilized to analyze transformers health state and assign corresponding health ratings to different clusters to guide the condition-based maintenance and operation dispatching. The feasibility and effectiveness of the proposed method are verified in case study. This work is supported by National High-tech R & D Program of China (863 Program) (No. 2015AA050204).