The dynamic modulus (|Em*|) of asphalt mixtures is one of most fundamental mechanical parameters used in flexible pavement design, which has been extensively estimated through various predictive models. However, most of existing dynamic modulus predictive models does not fit well for the purpose of the Level 3 (the lowest input level) flexible pavement design in the current AASHTOWare Pavement ME Design due to the following major deficiencies: (1) lacking physical significance; (2) providing poor prediction accuracy; and (3) requiring laboratory testing to obtain key model parameters. To address these deficiencies, this paper is aimed to develop a new model on the basis of the law of mixtures. First, a large database is collected and revised for the new model development. Two exiting law of mixture models are then evaluated in terms of the mode prediction accuracy, and neither of them provided satisfactory prediction results. A new model is subsequently developed, which involves using a parallel form to represent the contribution of each of three phases (asphalt, aggregate and air void) to the |Em*| as well as utilizing a standard sigmoid function to characterize the mechanical response of asphalt mixture. Statistical analysis shows that the new model has comparable prediction performance with traditional Witczak models, but with significantly lower number of input variables and relatively higher model quality. Supplemental data collected from laboratory tests and independent literature sources finally validate that the new model has the capability to provide an accurate prediction of |Em*| with an R2 value up to 0.90. In these regards, the new model is believed to be a good candidate for the |Em*| prediction of asphalt mixtures for a reliable Level-3 flexible pavement design.