Flexural beam fatigue testing of asphalt mixtures has been used for nearly 40 years in the pavement industry. Since the development of the test, the definitions of initial and failure stifnesses have not been verified or validated in any comprehensive study. The main objective of this study is to validate the criteria used to define the initial and the final stiffnesses in flexure fatigue testing. In this study, extensive flexure fatigue tests were performed on five typical dense-graded mixtures and an asphalt rubber gap-graded mixture. An optimization approach was used, in which different intial and failure conditions were assumed. Fatigue models were developed using linear regression curve fitting and the conditions that produced the best fit were selected. Both the phenomenological and the dissipated energy approaches were used. Test results conclusively indicated that the initial stiffness should be defined at cycle number 50. In addition, when a phenomenological approach for fatigue is employed, the fatigue stiffness should be taken at 50% of the initial stiffness.A stiffness degradation model was developed, which provided an independent proof that failue occurs when the stiffness of the beam is reduced to 50% of the initial stiffness. This model represents a basic material propertey at which damage accumulation in the mixture has produced an inability of the mix to resist further damage independent of the mode of loading. In contract to the tensile strain-failure approach, data analysis with the energy approach showed that fatigue failure stiffness, taken at 30% of the initial stiffness, provided identical fatigue energy failure regardless of constant stress or strain mode of loading. The results show that the phenomenological and energy approaches provide different definition of failure and the test should be consistent with the method of analysis used.