Traffic flow models are essential to performance analysis/evaluation for applications/services provided by Intelligent Transportation Systems (ITS)/Vehicular Ad-hoc networks (VANET). They are also useful guidance for the deployment of ITS/VANET. Many macroscopic traffic flow models have been proposed in the past few decades on the basis of empirical data collected in the US, Canada, Turkey and etc. However these models may not be accurate for traffic flows in cities in China due to the differences in population, transportation infrastructure, and driving culture. In this paper, we collected a large amount of empirical traffic flow data in Shanghai overhead road during three different time periods. Statistical results showed that the lane-level traffic volumes in Shanghai followed Gaussian distribution rather than Poisson distribution which was normally assumed in literature. Regarding lane-level vehicles' velocities, they matched well with Gaussian distribution. The empirical probability mass functions (PMF) for both the traffic volumes and vehicles' velocities were presented. In addition, how these models would impact performance analysis in VANETs was discussed.