This paper addresses the problem of freeway tra±c flow estimation. The freeway is considered as a network of components representing different freeway stretches called segments. The evolution of the traffic in a segment is modelled as a dynamic stochastic system, influenced by states of neighbour segments. Measurements are received only at boundaries between some segments and averaged within regular time intervals. An Unscented Kalman filter is developed and its performance is compared with a particle filter both for synthetic data and for real traffc data. The intended application is to supply traffc control systems with the estimated traffc state.