Abstract Fatigue devices working at high frequency (20 kHz) allow now to explore the fatigue domain with a total number of cycles much higher than 107 cycles in reasonable times. In that case, lifetime belongs to the very high cycle fatigue (VHCF) domain. As for low and high cycle fatigue (LCF and HCF), the fracture begins by microstructural damage because of microplasticity, before leading to the initiation stage (stage I), and then the short and long crack propagation stages (stage II). It is now admitted that in VHCF the initiation stage is higher than 90 % of the total life. The main difference with LCF and HCF is the crack initiation location. For materials without internal defects (type I), the initiation is always on the specimen surface, whereas for materials with internal defects (inclusions, porosities, etc.) named type II, the microstructure appears in subsurface on defects leading to a typical “fish eye” on fractographies. If the crack initiation mechanisms on surface are well accepted to be related to dislocations gliding in well-oriented grains, the mechanisms in subsurface are not very clear and many hypotheses on the fish-eye development are suggested by different authors. In the fish-eye center, it appears a fine granular area more or less large, the origin of which is debated. In the first part, this article proposes a review on these crack initiation mechanisms (surface and subsurface). Microplasticity produces a temperature increase in fatigue tests. It has been demonstrated that the temperature recording during tests is a mean to determine the number of cycles corresponding to crack initiation and crack propagation. In the second part, through two examples on Armco iron (surface initiation) and a low alloy steel (subsurface crack initiation), the method to determine the number of cycles during the initiation and propagation stages is exposed.