The performance issues and computation power consumptions of several prediction algorithms far real-time MPEG video are discussed on the paper. The real time MPEG traffic is most important application on broadband video application both on ATM or IP network. The characteristics of MPEG video traffic is burst, long-range dependent, and non-stationary with trend and periodicity. As the statistics of the underlying processes are either unavailable or uncertain in real-time applications, an effective prediction algorithm is helpful for user equipment to capture the real-time variant traffic. The time-invariant H2/H∞ filtering algorithm is suggested to estimate traffic parameters for long-range prediction of the MPEG traffic. Simulation results based on real MPEG traffic data show that the time-varying trend, the periodic components, and the long-range dependence property can be splendidly predicted and captured by the proposed method. Compared with the time-varying H∞, filter and the time delay neural network (TDNN), the proposed scheme has the best compromise between prediction performance and computation cost for long-range prediction of the MPEG video traffic.