Power customers load profile clustering is the basis for constructing an appropriate tariff system and applying load management measures. This paper is concerned with the load profile clustering of low-voltage terminal customers using the self-organizing map（SOM） neural network. Three types of vectors， namely， the power curve， the time sharing power and the power spectrum are defined and used as inputs of the SOM neural network to realize clustering visualization. Two indices， the relative quantization error and topology error， are introduced to evaluate the clustering quality. After the assessment， the SOM output layer with optimal performance is selected to classify load profiles with the k -means method. The 131 profiles concerned are classified into eight types according to Davies index， with each group of profiles described. The new customer identifying ability of the SOM neural network is examined. The result shows that the method proposed is effective and reliable.