Random vibration excitation is a common cause of failure, especially when the vibration is in the range of natural frequencies, where the stress response is greatly amplified. A vibration-fatigue analysis can be performed for a high-cycle regime consisting of a structural dynamics analysis, a response calculation and a fatigue analysis. The material parameters (S-N curve) are defined for a constant-amplitude, cyclical, uniaxial stress state. However, in real structures the stress state due to the structural dynamics is rarely uniaxial the and direct application of the S-N curve is difficult. The stress tensor is reduced to a more manageable representation using a multiaxial criterion. A multitude of such criteria are available in the literature. In this study, a group of multiaxial criteria are compared theoretically and experimentally, i.e., maximum normal stress, maximum shear stress, maximum normal-and-shear stress, C-S criterion, Projection-by-Projection and the Preumont and Piéfort criterion. A special specimen is used in the experiments that experiences a rich structural response which causes fatigue failure. The experimental comparison of the crack location and the time-to-failure gives comparable results for the tested multiaxial criteria with a reliable time-to-failure estimation. From the research it follows that the crack-location estimation is not accurate enough for either of the compared criteria. The study proves the applicability of the vibration-fatigue analysis procedure (i.e., from excitation, structural dynamics, multiaxial criteria to spectral moment methods) on real vibrating structures with rich structural dynamics.