The headway between vehicles in a traffic stream is of fundamental importance in traffic engineering applications. Previous research in this subject has focused on modeling theoretical distributions for low and medium traffic flow conditions. Yet little research has studied congested traffic conditions i.e., the high traffic flow state. In the same context, there appears to be a lack of clear-cut boundaries for the three flow states (low, medium, and high). This study attempts to determine such boundaries on the basis of traffic conditions observed at the study sites. Although observed headways at arterial sites follow a lognormal distribution, distributions that fit closure headways differ according to the traffic flow state. The Erlang distribution provided a good fit to the observed headways at sites with high traffic flows constituting different modes in traffic composition. Two population model and three population models were observed at high flows. The low volume level indicated the fit of the lognormal distribution.