Three-dimensional excitation and emission matrix fluorescence spectra (3D-EEM) has attracted the increasing attention of researchers in water monitoring and water treatment areas. The self-organizing map (SOM) is a kind of non-supervised and self-learning neural network with the feature of high self-stability and noise tolerance. In the present paper, SOM technique was employed for the exploratory analysis of EEM spectra of water samples in a water treatment plant. The results showed that EEM spectra could be clustered into three classes, corresponding to tryptophan-like protein substances, tyrosine-like protein substances, and UV fulvic-like substances. The three components could be effectively removed during the whole water treatment process with the high removal of 84.6% (tyrosine-like), 79.9% (tryptophan-like), and 69.1% (UV fulvic-like). The results show that SOM technique can be used as an effective tool for EEM spectra analysis, which is helpful for the optimization of water treatment process parameters, the improvement of process performance, and the operation of water treatment plant.