Abstract: In general, water quality time series not only has stationary and periodicity characteristics, but also has obvious multi-scale features. To improve the precision of the traditional autoregressive models, which were once widely used for forecasting water quality, the autoregressive model combined with the multi-scale wavelet analysis theory is proposed as a new forecasting model called WAR(Wavelet Autoregressive Model ). Finally, this new method and the traditional autoregressive model are applied to predict four water quality indicators in the Lianghui reservoir. The analysis results show that the WAR model has significantly improved the prediction accuracy in comparison with the traditional autoregressive model. At the same time it is also found that the model is feasible and practical, and can only provide references for similar studies. In view of this, when we predict water quality, in order to improve the prediction accuracy, it is important to choose the model according to actual situations, and this point is crucial.