In order to establish the predictive model adapt to the variation law of groundwater resources in the region of the Dongting Lake, on the basis of analyzing the relativity of nature river flow, Yangtze three inlet inflow water and Chenglingji outflow water with water resources, the paper respectively used multi-factor stepwise regression model, back-propagation neural network model and multivariate time series CAR model to set up three kinds of predictive model of groundwater resources in Dongting Lake area and compared and analyzed the overall law of prediction accuracy and prediction results of the three models.The results showed that the amount of groundwater resources have better relativity with the natural river flow, Yangtze three inlet inflow water resources, and Chenglingji outflow water; the prediction accuracy of CAR model of multivariate time series is better,and then that of Back-Propagation Neural Network is the second, and the prediction accuracy of multi-factor stepwise regression model is poor; the overall law of the prediction results of multivariate time series CAR model is better than that of back-propagation neural network model, and the overall law of prediction result of Back-Propagation neural network model is superior to that of multi-factor stepwise regression model.