This paper presents a neural network approach intended to aid electricity consumers or power system planner in the task of classification of harmonic load types using the measurements (data samples) collected from one of the power station in Malaysia. In order to allow the classification of the type of harmonic loads, harmonic currents order produced by aggregate harmonic loads and the level of emission are modelled using the Multi-Layer Extreme Learning Machine with autoencoder (hereinafter denoted as ML-ELM-AE). The feasibility of ML-ELM-AE on the classification of the Harmonic empirical data set will be probed, for when the classification results of the Harmonic load types become available, it can come in handy to power system analyst or engineers for analysis in later stage. They can use it to determine the harmonic distortion patterns and to characterize the harmonic currents at network buses. Depending on whether the aggregate load is residential, commercial, or industrial, the load characteristics in terms of its harmonic contents are likely to be different. The achieved results demonstrate the effectiveness of the investigated technique in dealing with the real world power system application of harmonic load type classification, that can be useful in providing good indication to the power system or distribution network planner.