Discrete-time stochastic models of sequential traffic records in telephone networks are derived from classical assumptions of queueing theory. They appear to be well suited for a use of Kalman filter. Thus, a version of the filtering algorithm is adopted to deal with the traffic fluctuations. The overall approach is tested against the real data gathered by the supervisor of the French toll network. Next, the adaptive traffic estimator is inserted into the simulator of a testbed network operated in the urban zone of Paris. The traffic estimations provided by the filter are used to decide the link overflow hierarchy.