Conventional economic models of traffic congestion assume that therelation between road use and speed is a technical one. In this paper wederive the speed-flow relationship from more fundamental considerationsconcerning driving behaviour. We develop a structural model in which driverschoose their own optimal speed, by trading off various cost aspects of making atrip: time costs, expected accident costs and fuel costs. Since the optimalspeed depends on the presence of other drivers on the road, we can derive a speed flow relationship on the basis of this behaviour. It is demonstrated thatthe relationship between the various cost components should be taken intoaccount in computing the external costs of traffic. For tolls alone, it isdemonstrated that a regulator ignoring the fundamental relation willotherwise fail to set optimal tolls, and will underestimate the efficiencygains of congestion pricing. Moreover, the overall welfare optimum in ourmodel is found to be off the speed-flow function, and off the average andmarginal cost functions derived from it in the conventional approach. This fulloptimum requires tolls to be either accompanied by speed policies, or to beset as a function of speed. Using an empirically calibrated numericalsimulation model, we illustrate these qualitative findings, and attempt to assess their potential empirical relevance.