Logistics demand forecasting is important for investment decision-making of infrastructure and strategy programming of the logistics industry. In this paper, a hybrid method which combines the Grey Model, artificial neural networks and other techniques in both learning and analyzing phases is proposed to improve the precision and reliability of forecasting. After establishing a learning model GNNM(1,8) for road logistics demand forecasting, we chose road freight volume as target value and other economic indicators, i.e. GDP, production value of primary industry, total industrial output value, outcomes of tertiary industry, retail sale of social consumer goods, disposable personal income, and total foreign trade value as the seven key influencing factors for logistics demand. Actual data sequences of the province of Zhejiang from years 1986 to 2008 were collected as training and test-proof samples. By comparing the forecasting results, it turns out that GNNM(1,8) is an appropriate forecasting method to yield higher accuracy and lower mean absolute percentage errors than other individual models for short-term logistics demand forecasting.