In order to enhance the prediction accuracy of water demand in Donghai County, this paper extracted irrigation water demand quantity which counted high proportion and showed strong randomness from the total water demand. Quota analysis method was taken to predict irrigation water demand quantity on the basis of different frequency years. Compound prediction method composed with the smooth model SM2and the grey prediction model was taken to predict non-irrigation water demand. When the weight of each single model was set, the impacts in residual errors in recent periods on the prediction value were more considered. Final prediction could be available by summating the quantity of irrigation and non-irrigation water demand. This method requires less data and can give ranges of water demand quantity based on different probabilities. The final prediction result was reasonable judged by the history data of water demand in Donghai County.