In order to better study the car-following characteristics of vehicles in complex situations, the car-following behavior of vehicles is analogous to the result of the interaction of molecules in a one-dimensional pipeline. The existing car-following model based on molecular dynamics uses the demand safety interval and the expected vehicle speed as the main factors to establish the molecular car-following model, ignoring the influence of the relative speed on the driver’s car-following behavior, which does not conform to the real car-following situations. Therefore, by incorporating the relative velocity of the vehicle into the model structure, this paper establishes the molecular interaction potential function and the wall potential function, and constructs a car-following model with improved molecular dynamics accordingly. The parameters in the car-following process are collected by high-precision vehicle-mounted instruments and the model parameters are calibrated by genetic algorithm. Finally, the accuracy of the model under different car-following states is verified and compared with the previous molecular model. The results show that the improved molecular car-following model can more effectively predict the car-following behavior of the vehicles.