It is important to evaluate power supply service quality objectively and then to locate weak points accurately to improve the power supply service level for clients. Given this background, a client-side power supply service evaluation index system is constructed from the aspects of customer demand, responsibility attribution, work order irregular change and service processing. An index weight determination method based on an entropy weight method and a correction factor is proposed to overcome any unreasonable weight distribution in the case of similar distrbution of index value; a comprehensive evaluation method of power supply service quality based on an improved Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) and cosine similarity degree is proposed. This comprehensively meets the requirements of power supply service quality. Then, a method of power supply service quality grade evaluation based on bi-level clustering and fuzzy theory is proposed, in which the clustering results are used as the basis of grade evaluation. This is to overcome the shortcomings of strong subjectivity of grade evaluation in power supply service. Finally, a case of the service quality of power companies in Zhejiang, China is studied. The results show that the client-side power supply service evaluation method proposed in this paper can effectively evaluate the service quality level of each power company, and provide references for service rectification. This work is supported by the National Key Research and Development Program of China (No. 2016YFB0901100).