Reliable predictions on import prices of agricultural products can result in appropriate timing of import and saving of exchange resources. Linear time series models including ARIMA, GARCH and EGARCH have frequently been used in time series forecasting during the past three decades. Recent studies on forecasting with artificial neural network (ANNs) suggest that ANNs can be a suitable alternative to the traditional linear models. But neither linear time series models nor ANNs can be adequate in modeling and forecasting time series since the linear model cannot deal with nonlinear relationships while neural network model alone is not able to handle both linear and nonlinear patterns equally well. Hence by integrating linear time series with ANN models and designing the hybrid model, patterns in the data can be explained more accurately. In this research, a hybrid. methodology that combines time series model ARIMA, GARCH and EGARCH and ANN models is designed and results are compared with those of competitive models. In order to compare forecasting accuracy, in addition to the usual criteria such as RMSE, MAD, MAPE and Theil Coefficient with introducing Granger and Newbold statistic, significance of forecasting accuracy difference have been investigated. Main finding for price of three import products including Wheat, Corn and Sugar indicate that hybrid model significantly improves forecasting accuracy. In order to save more exchange, application of hybrid models for agricultural price prediction, especially for strategic import products, is recommended.