The new developments in power system such as restructuring and competitive electricity market make power quality (PQ) an important factor in competition. However, finding a measure for PQ evaluation is very difficult due to many indices involved in PQ measurement. For this reason, obtaining a single quantitative index based on the standard measurements has been a new challenge in recent researches. In this paper, a data mining method is proposed to determine global indices for PQ. The continuous and discrete indices of PQ are considered and a Unified Power Quality Index (UPQI) is presented for each PQ index, based on the method of incorporation and normalization. The indices are normalized and classified. Then, the global PQ index of each distribution site is determined by the Fast Independent Component Analysis (FICA) algorithm. In this approach, the PQ measurements of 313 real distribution sites are used to assess and classify the indices for different type of loads in the real distribution system. The results show the capability of this method to obtain an accurate measure for PQ evaluation. In this method, the convergence rate is very fast. Also for evaluating the accuracy of the proposed algorithm, an intelligent method based on artificial neural network (ANN) and fuzzy logic to obtain a global index for PQ assessment are implemented that the comparing between this two method show that the proposed method is stronger and proper than the intelligent method. This method can be extended for many distribution sites.