Abstract: For describing the distribution characteristics of headway, driving behaviors are classified into car-following state and free driving state by reference to the two-fluid theory. After analysis of their operation features, a mixed distribution model that can describe the headway distribution of both the two driving states was built. The model parameters were determined by expectation maximization (EM) algorithm using the measured data. Then, the differences between the statistical characteristics of the driving behaviors in up-stream section, downstream section and different lanes were analyzed and verified through a case study. The results show that the mixed distribution model can pass the chi-square test in all road situations. Compared with the negative exponential distribution, Erlang distribution, and M3 distribution, the mixed distribution model can improve the fitting accuracy of simulation by more than 10%, and obtain a more close result to the measured data in calculating the traffic capacity of an expressway on-ramp.