The second part of the study adopts the seawater salinization factor and arsenic pollutant factor to forecast the variation of ground water quality in Yun-Lin coastal area. Back-Propagation (BP) neural network which has the characteristics of self-organizing, self-learning and nonlinearity is selected to forecast future variation of groundwater quality. The influence of hidden nodes to the water quality forecasting is discussed first, the accuracy of the water quality forecasting results using different BP network input model are also analyzed. The results show that the hidden nodes are not a significant factor to BP network training and forecasting. Using recent variations data with high relativity in the input layer gives better results on network forecasting. Besides, the confident intervals of each forecasting value are also computed. The results indicate that the neural network is capable to describe the complex variation of groundwater quality and provide good forecasting reliability.