The conventional method of power user clustering only considers physical and demand side information such as the amount and the peak and valley difference of power consumption. It is not appropriate in a new era where wholesale and retail power markets have been liberalized and power system has becoming increasingly intelligent. The new method developed in this study introduces price signals at both wholesale and retail power markets into the power user clustering and considers both business value and demand response value of power users on the basis of metric learning algorithm (MLA). The case study shows that this new method significantly improves the degree of separation between the business value and demand response value indicators of clusters and reveals the relationship between the wholesale power price and the weight of power consumption for a specific target. This new method can help power retailers in their business decision making and has strong applicability and expandability.