This paper presents a traffic prediction method for MPEG-4 (Moving Picture Experts Group) VBR (Variable Bit Rate) videos. No previous knowledge of the traffic statistics is needed when Kalman optimized method is applied to modeling and predicting MPEG-4 video traffic. The GoP (group of pictures) loss probabilities in actual traffic are a very close approximation to those in our prediction traffic and the prediction errors are very small. Simulations show our prediction method for future GoP is effective and applicable to real time video traffic prediction and resource reservation.