Abstract: To improve the accuracy of tidal prediction, this paper presents a time series model using artificial neural network combined with chaos theory, and this model has been developed to overcome the limitation of empirical model and traditional neural network model. It determines the existence of chaotic behaviors in the data series of every-other-day difference in tidal time; then, phase-space reconstruction for the error series of empirical model is applied to neural network inputs. The model can give a prediction of errors that is useful for modifying or updating the final results. Prediction of the tidal times during one month at four tide observation stations on the Qiantang River shows that the model reduces the root mean square error (RMSE) by 83.9% and has accuracy higher than the traditional model.