The study of power quality (PQ) warning threshold is for the development of PQ early warning system. In this paper, a method of determining the PQ threshold based on k-medoids clustering is proposed to solve the problem that the current segmentation method of PQ warning threshold of is a complex and single. The method firstly analyzes the PQ data by cluster analysis, then uses square sum error based on distance as the clustering quality metric function. According to the actual situation determined by the threshold, the method sets the value of k as 2, naturally divides all the data into two classes: normal class and abnormal class, on the previous basis, the threshold is selected. The experimental results show that, for the selection of PQ threshold, the above method has good effect and efficiency.