Traffic simulation has been widely used for the evaluation of the traffic environmental impact, where the simulation model outputs are linked with energy/emission models. However, significant differences have been noted between field measurements of traffic operations and speed/acceleration predictions from car-following models which significantly affect energy consumption and emission predictions. Carfollowing models were originally developed from (and calibrated with) aggregate parameters, while fuel consumption and emissions models require instantaneous vehicle activity as inputs. In this research, widely used car-following models are re-assessed for use in emission/fuel consumption estimation, with their impact on vehicle specific power and instantaneous engine work as comparative indices. Second-by-second vehicle trajectories are used in these comparative analyses. Simulated VSP distributions and instantaneous engine work are compared with field distributions to verify these models. A new car-following model based on machine learning is proposed for more accurate estimation of emission/fuel consumption.