We report on the analysis of the residual chlorine in the water distribution system. A neural network model is used to forecast day to day variation of chlorine density in the water at various places. The network is trained with one-year data of water temperature, precipitation, original chlorine density at the water supply center, and residual chlorine density at nine measuring places. It well reproduces the residual chlorine density of the given data and can predict day to day variation in other years.