The effect of high impedance faults (HIFs) in a distribution system has an effect on the system stability and reliability. The high impedance arc fault (HIAF) even poses a significant threat to the living being as it involves arcing. The HIAF have the same V-I characteristics as HIF, but the enormous amount of heat generation is a significant concern to differentiate between them. Moreover, there are different types of arc which may occur depending on the arcing conditions and arcing surfaces. In this article, arc due to leaning-tree on a medium-voltage distribution line and the arc in between sphere-gaps are considered as the main event of analysis. An empirical mode decomposition (EMD)-based approach is used along with artificial neural network (ANN) for the analysis of real-time arc signals. The result obtained by the application of EMD and ANN on arc voltage signals successfully detects and classifies the arcing events by their predominant harmonic signatures. A comparison of EMD results with other techniques is presented. Arc in soil and wet-sand is also considered for the validation of the proposed algorithm for other arcing surfaces. Support vector machine and Kth nearest neighbor machine learning techniques are also utilized for classification. © 2007-2012 IEEE.