The development of statistical theories of traffic flow on highways has often started with a model in which each driver retains a fixed desired speed. This has been taken to mean that the distribution of desired speeds for the cars on the road at a given time was independent of density. However, it is shown that this model really implies that the desired speed distribution for the cars crossing any fixed point on the road in a certain period of time will be density-independent. The relation between the two distributions is computed, and the effect of the distinction on a theory of the density dependence of traffic behavior is discussed. The important input for a statistical theory of traffic flow seems to be a description of the desired speeds of the cars using the road in a period of time and a model of passing and queuing behavior. The number of cars crossing a fixed point in unit time appears to be more basic in such a theory than the density of cars.