Fatigue crack is one of the most common distresses in asphalt pavement. In this study, the fatigue crack behaviors of asphalt binders were evaluated under repeated cyclic controlled-stress loadings. Fatigue crack density, defined by the initial sample radius and effective radius of the undamaged part during fatigue testing, was used as an indicator. Also, an effective radius function was developed using the torque equilibrium principle and dissipated strain energy (DSE) equilibrium principle. The data revealed that fatigue crack density of asphalt binder manifested three changing tendencies, in which the secant method was employed to partition their profiles based on concavity and convexity characteristics. Specimens made of #50, #70 unmodified asphalt, SBS modified asphalt, and #50 asphalt mastic, respectively, were tested by time sweep tests on a Dynamic Shear Rheometer (DSR) at different stress levels and rest time. The variations in fatigue crack density at the three stages, as well as the changes in fatigue crack density growth rate in the second stage, were all analyzed. The asphalt binder showed fatigue crack density profiles composed of three stages: crack initiation, steady growth, and unstable growth. Stage II had the smallest increase in fatigue crack density among the three stages but underwent more than 50% load cycles throughout the whole fatigue process. Hence, fatigue crack density growth rate (RD) could be used to evaluate the fatigue resistance of asphalt binders. A positive power-law relationship between RD and stress level was recorded since higher stress levels meant more serious damage. By contrast, RD displayed a negative power-law relationship with resting time, revealing that the rest time, that is, self-healing, effectively postponed the fatigue crack propagation rate.