In this study, the applicabilities and validities of neural network approaches to the prediction of THM formation in a water distribution systems were analysed and discussed. The estimated THM formations by multiple linear and exponential models were compared with the prediction results of neural network model based on the proposed variables by AWWA. The model equations of THM formation were determined by multiple exponential and linear regression through statistical analysis of the observed data set. Input parameters of neural network models for the prediction of THM formation were consisted with organic compounds observed in water distribution systems such as TOC, DOC and UV254. As results, it was shown that the neural network models were more efficient and robust method for the prediction of water quality in distribution systems. From the results, it can be concluded that the proposed methods are predicted well the characteristics of observed THM formation.