Hydrologic time series have more dynamic and nonlinear characteristics of system. This study attempted to simulate the daily stream discharge of Chien-Feng Bridge with the dry and wet cyclical phenomena systematic variation characteristics by utilizing nonlinear time series, i.e. Threshold autoregressive model (TAR) and Bilinear time series model (BL) were built by genetic algorithm to seek the optimum orders and parameters of models. Besides, several linear models, including ARIMA and TFN models, were also constructed for comparison. The analytic results show that the threshold autoregressive model has the best forecasting ability of all models.