In packet switching network such as asynchronous transfer mode (ATM), the switching characteristics is important in delivering the guaranteed QoS (Quality of Service) level of the network. Many methods have been developed to control cell flow for shared bandwidth. The first-in first-out (FIFO), static priority (SPR), dynamically weighted priority scheduling (DWPS) (T. Lizambri, F. Duran, and S. Wakid, 1999) and weighted fair queuing (WFQ) (R. Händel, M.N. Huber, and S. Schröder, c1998) are some of the schemes for managing the shared bandwidth. Due to the diversity of services supported in ATM network, it is typical for the traffic flow pattern to change dramatically. A common trait of these algorithms is that their mechanisms are fixed, and they cannot adapt efficiently for such traffic flow changes. In order to address this, we propose an evolutionary fuzzy system (EFS) scheme to do ATM cell scheduling. With EFS, the fuzzy switching algorithm can be adjusted to track the changes in the pattern of traffic flow in order to maintain the desired level of performance. The desired quality of service (QoS) performance level can be conveniently achieved by tuning the parameters of the fitness function.