Guazhou oasis locates in downstream of Shulehe River Basin, belongs to national extreme arid desert nature reserve, and the formation of its water resources is difficult, and the ecological environment is fragile. BP neural network model which based on the time series and addition of the momentum term was used to model and predict the annual average of water quality indicators in Shuangta Reservoir from 1993to 2008. And the parameters of the model are the followings, the number of the enter nodes is 2, the number of the output nodes is 1, the number of the hidden layers is 2, minimum training rate is 0.1, dynamic parameters is 0.6, adjusted value of the SIGMOID function is 0.9, permissible error is 0.0001, maximum number of iterations is 10000. Moreover, the value of the relative error of model fitting is less than 5%, and the value of the relative error in the stage of prediction testing is less than 10%. According to the results of the water prediction, the water quality from 2009to 2013is belong to Class II grade which meets the standards of domestic and agricultural irrigation water use, however, the water has the risk of eutrophication.