Mechanical components, such as gears, are usually subjected to variable loads that induce multiaxial non-proportional stress states, which in turn can lead to failure due to fatigue. However, the material properties are usually available in the forms of bending or shear fatigue limits. Multiaxial fatigue criteria can be used to bridge the gap between the available data and the actual loading conditions. However, different criteria could lead to different results. The main goal of this paper is to evaluate the accuracy of different criteria applied to real mechanical components. With respect to this, five different criteria based on the critical plane concept (i.e., Findley, Matake, McDiarmid, Papadopoulos, and Susmel) have been investigated. These criteria were selected because they not only assess the level of damage, but also predict the direction of crack propagation just after nucleation. Therefore, measurements (crack position and direction) on different fractured gear samples tested via Single Tooth Bending Fatigue (STBF) tests on two gear geometries were used as reference. The STBF configuration was numerically simulated via Finite Elements (FE) analyses. The results of FE were elaborated based on the above-mentioned criteria. The numerical results were compared with the experimental ones. The result of the comparison showed that all the fatigue criteria agree in identifying the most critical point. The Findley and Papadopulus criteria proved to be the most accurate in estimating the level of damage. The Susmel criterion turns out to be the most conservative one. With respect to the identification of the direction of early propagation of the crack, the Findley criterion revealed the most appropriate.