Network characterization and modelling is an important issue to understand and monitor IP network performance, in particular for real-time multimedia applications. To maintain an adequate quality of experience for end customers, we need to monitor and eventually control the effects of the network behavior. Thus, an accurate network performance model is needed, which is not a simple task, given the complex dynamics of a network. In this paper, we propose to represent the effect of an IP network on a video connection as a single G/M/1 queue. When a video session is set up, the two ends can infer the queue parameters based only on local measurements of the stream itself and with a single exchange of information between the two ends. We found that the recommended queue parameters may be different from one connection to the next but that the actual shape of the arrival process does not have a large impact on the selected parameters. In fact, using a lognormal distribution may work for many cases. Finally, we found that jitter can be used as a proxy to estimate network delay without the need for end-to-end synchronization.