In recent years, growing processing power and bandwidth have stimulated a new breed of network applications requiring the development of more sophisticated traffic management algorithms to meet their stringent QoS needs. One of the main obstacles to improving QoS is congestion at network nodes. It is believe that bandwidth prediction is a key component to reducing congestion and thereby the improvement of QoS. This work proposes an effective bandwidth management approach based on the discrete wavelet transform (DWT). The approach hinges on the results of recent studies showing that while network traffic consists of both long- and short-term dependence, its wavelet coefficients are short-term dependent only. The DWT decomposes the traffic into a low-frequency component and high frequency components. The low- frequency component of the traffic, are significant for long-term traffic behavior and, hence, for bandwidth allocation. The high-frequency components are significant in the short-term behavior of the traffic and, hence, for buffer-allocation.