The rapid increase in computer technology and the availability of large scale power quality monitoring data should now motivate distribution network service providers to attempt to extract information that may otherwise remain hidden within the recorded data. Such information may be critical for identification and diagnoses of power quality disturbance problems, prediction of system abnormalities or failure, and alarming of critical system situations. Data mining tools are an obvious candidate for assisting in such analysis of large scale power quality monitoring data. This paper describes a method of applying unsupervised and supervised learning strategies of data mining in power quality data analysis. Firstly underlying classes in harmonic data from medium and low voltage (MV/LV) distribution systems were identified using clustering. Secondly the link analysis is used to merge the obtained clusters into supergroups. The characteristics of these super-groups are discovered using various algorithms for classification techniques. Finally the a priori algorithm of association rules is used to find the correlation between the harmonic currents and voltages at different sites (substation, residential, commercial and industrial) for the interconnected supergroups.