The mechanical design of structural components for high cycle fatigue applications needs reliable fatigue limits. However, mainly because of notches and the unavoidable presence of small defects, such a task still presents some challenges. Notches cause a stress concentration effect that can initiate short cracks at their tips, but such short cracks may propagate or become non-propagating, depending not only on the load level, but on the stress gradient ahead of the notch tip as well. Notch-like defects, such as scratches, pores, and inclusions, behave in the same way. There are empirical and theoretical models to predict the fatigue limit of notched components. The latter includes the so-called Stress Gradient (SG) model, based on linear elastic fracture mechanics concepts and using the El Haddad-Topper-Smith (ETS) characteristic size aR, as a promissory approach. However, there is a lack of experimental data verifying their fatigue limit predictions. In this context, C(T)-like notched specimens of SAE 1020 steel with several notch root radii were tested under constant load amplitude control at 40 Hz and a stress ratio R equal 0.1, to evaluate their fatigue limit through accelerated tests involving step loading procedures with blocks of 3.10 to sixth power cycles. The experimental fatigue limit was compared with values predicted by SG model, following three approaches: SGc-p, SGs-e, and SGquebra, according to the determination of the geometric factor of the stress concentration factor; and with an alternative prediction by the Point Method based on the theory of critical-distance (TCD). SGc-p, SGquebra and TCD model predictions are almost coincident for blunt notches and they present a good agreement with experimental results, but they are non-conservatives in the case of sharp notches; while SGc-p predictions are conservative for both blunt and sharp notches. Since both models are based on linear elastic concepts, it was demonstrated that an elastic analysis presents limitations to model the behavior of short cracks emanating from sharp notches, due to the local stress at the critical point can exceed the yield strength of the material. Furthermore, according to SG model, the fatigue limit is related to the presence of non-propagating short cracks (NPSC). Such surface NPSCs on the face of the specimens were monitored by non-destructive techniques including optical microscopy, digital image correlation (DIC) and micro-computed tomography; whereas subsurface NPSCs were detected through destructive metallographic technique. The sizes of the detected NPSCs were much smaller than those values predicted by SG model, which in turn makes the detection of these cracks a more complex problem.