Under fatigue loading, the number of cycles to failure and its associated scatter increase when the loading level decreases. The High-Cycle Fatigue (HCF) regime is thus characterized by a large scatter in the number of cycles to failure . Cracks initiation represents an important part of the lifetime of the structures. A stochastic method is used to study the fatigue crack initiation prediction in the 316L austenitic stainless steel. The present work proposes to show that this scatter can be attributed to the random orientation of individual grains, which influences the crack initiation localization. The stresses in grains are determined by finite element computations (FEM ), using a configuration representative of a polycrystalline aggregate. This approach takes into account the crystallographic orientations of the grains in the aggregate as well as the deformation incompatibilities between neighbouring grains due to crystalline anisotropic elasticity and elasticplasticity . Then, the scatter of the number of cycles to crack initiation is derived from the FEM stress fields using two fatigue crack initiation criteria: an usual one, Mura’s criterion  and a more recent one , based on Discrete Dislocation Dynamics (DDD) simulations and taking into account plastic slips, cross slip and stress tensor components.