To study the fatigue failure model of steel round bars, the unstable propagation area, stable propagation area, and stable propagation length of fatigue crack were calculated based on the fracture criterion from an elliptical fracture model of steel and the true stress field at tip of fatigue crack in round bars. On the basis of the facts that fatigue crack propagation accelerated with the numbers of cyclic loading during fatigue tests, it was assumed that the stable propagation rate of fatigue crack and the numbers of cyclic loading of constructional steel round bars comply with a monotonically increasing power function. In other words, the stable propagation rate of fatigue crack and the numbers of cyclic loading of constructional steel round bars are monotonically increasing linear functions in double logarithmic coordinate system. A function expression between the stable propagation length of fatigue crack and the fatigue life of constructional steel round bars (i.e., the fatigue failure model of steel round bars) was derived by integrating the power function between the stable propagation rate of fatigue crack and the numbers of cyclic loading. The proposed fatigue failure model of steel round bars shows that the fatigue life is a complex function of nominal maximum stress, relative stress amplitude, and location and length of initial crack, which cannot be merely simplified as a function of stress amplitude. Fatigue tests under constant amplitude cyclic stress were carried out on Q345B steel round bars . The fatigue test results show that the fatigue life of Q345B steel round bars decreased with the increase of relative stress amplitude and nominal maximum stress. The fatigue model parameters were calibrated and the model accuracy was verified, according to fatigue test results of Q345B steel round bars .