This paper presents a combined forecasting model for uncertain demands based on rough set (RS) and radial basic function (RBF) network. First, RBF network is introduced to work on the historical data and extrapolate future demands. Then attributes reduction of the RS is applied to analyze the datasheet and draw the kernel index set from it. The forecasting results are obtained with a combination of the above two methods. The combined model is tested with real historical data collected from a large firm in the automobile industry, and it produces more precise results than the RBF model