There are many ways to predict drinking water quality such as neural network, gray model, ARIMA. But the prediction precise is need to improve. This paper proposes a new forecast method according the characteristic of drinking water quality and the evidence showed that the prediction is effectively. So it is able to being used in actual prediction.