By approximating a packet network, such as the Internet, by an M/M/1 queuing model, the available bandwidth of the network can determined from the variance of the distribution of the sojourn times of packets transmitted through the network. In order to determine the variance of the distribution of sojourn times, and thus the bandwidth of the packet network that is available for transport of packets from an input to an output, a probe train of N packets with known inter-packet intervals is injected from an input on a sending side of the network. At the receiving side of the network, the packet arrival times are measured and the inter-packet intervals of each pair of received packets in the received train are calculated and compared with the known inter-packet interval times of the corresponding input packets. The variance of the distribution of the sojourn times of the packets through the network can then calculated using these time differences even though the individual sojourn times of the packets through the network cannot be directly measured.