Abstract-In this paper, the flow-level performance of opportunistic scheduling in orthogonal frequency division multiplexing (OFDM) networks is studied. The analysis accounts for the applications with a dynamic number of competing flows, such as continuous transfers of file transport protocol (FTP) or web browsing sessions. An analytical model is developed to extend the multi-class Processor-Sharing model in single-carrier networks to multi-carrier OFDM networks, where the total service rate varies with the number of flows. Based on the analytical model, the scheduling gains in both OFDM-TDMA (time division multiple access) and OFDMA (orthogonal frequency division multiple access) networks are evaluated for low and moderate signal-tonoise ratio (SNR). Different from previous works, we focus on the scheduling performance at the flow level and consider a dynamic network setting with random sized service demands. Furthermore, we use stochastic comparison techniques to examine the effects of physical-layer characteristics, such as fading speed and channel frequency selectivity, on flow-level performance. Simulations are performed to verify the analytical results.