The forecasting of water quality variation is very important in the process of sewage treatment, which helps the control system work reliably and steadily. In this paper, the compensative fuzzy neural network (CFNN) based on compensative fuzzy logic and neural network and its study arithmetic are introduced. Considering its features as fast speed, steady studying course, global dynamic optimization, CFNN is applied to establish water quality forecasting model. The practical example indicates that the model is not sensitive to initial parameters and has better forecasting precision and faster convergence.