This work aims to evaluate the suitability of multiaxial fatigue models regarding their application to crankshaft shafts, as well as evaluating the possibility of crankshaft failure when subjected to loading conditions calculated for the operation. Fatigue is a mechanical failure primarily caused by the application of time-varying loads, whose main characteristic is the generation and/or propagation of cracks. It is extremely important that the study and modeling of fatigue failures is accurate and reliable, since the generation and propagation of cracks, in general, do not cause an evident change in the overall behavior of the structure. This is because the damage generated by a crack is, almost always, restricted to the critical region of the part, tending not to generate early warnings of imminent failure. Fatigue failure predictions become critical for both uniaxial and multiaxial loads. However, for multiaxial loading conditions, analyzes based on more advanced models than those adopted for uniaxial conditions are required, since their stress state is extremely complex. In this context, criteria were developed for an attempt to predict component failures subjected to cyclic multiaxial loadings. An example of a mechanical component subject to multiaxial fatigue, when in service, is the crankshafts of power plant generating units. Thus, the present work analyzed the applicability of seven models (Findley, Matake, McDiarmid, Susmel and Lazzarin, Papadopoulos, Carpinteri ans Spagnoli and Liu and Mahadevan) of high cycle multiaxial fatigue, based on the critical plane, for failure prediction in crankshaft shafts of generating units in thermoelectric power stations. For the application of the models, loading conditions reported in the literature as critical for failure and material properties (fatigue strength limits) previously obtained experimentally will be used as input data and, from that, an Error Index will be generated for each of the seven models evaluated. Differences were observed between the results for each model, with some being more conservative than others, precisely due to the loading conditions and material properties used. Despite this, the Papadopoulos model proved to be the most suitable among the seven models, due to its ease of formula application and precision in predicting failure.