In this paper, a data mining technique is proposed for distribution network voltage regulation. The proposed method efficiently extracts rules for dealing with three-phase distribution network loss minimization. Distribution companies are concerned with the network operation to make a profit under competitive environment. From a standpoint of power quality, the mathematical formulation of distribution network voltage control is hard to solve due to the complexity. In that sense, distribution operators do not understand the problem features well. This paper presents a regression-tree-based method that clarifies the nonlinear relationship between control variables and the network loss in a distribution network. A regression tree is constructed to find out rules in a distribution network voltage regulation. The proposed method is successfully applied to the IEEE 34-node system.