In a packet voice system, discarding of a packet is performed as a function of previously discarded packets. In one embodiment, a packet voice system includes an ATM Adaptation Layer Type 2 (AAL2) and Service Specific Convergence Sublayer (SSCS) System. In this system, a transmission buffer stores AAL2 voice packets for transmission, each AAL2 voice packet comprising a sequence number, the values of which range from 0 to n−1, and a source identifier, k. When traffic congestion is detected, the transmitter portion of the SSCS System selectively discards one packet from a source k at the output of the transmission buffer if no packet from source k was dropped in either the last n−1 packets or over a predefined prior interval of time. Another embodiment of the invention discards packets at the input of the transmission buffer.