The lifetime to initiate an engineering crack is usually predicted from S–N curves in high cycle fatigue or Coffin–Manson curves in low-cycle fatigue. This paper investigates how to predict the engineering life to initiate an engineering crack from the growth of small cracks. Small cracks can nucleate from defects and especially ceramic inclusions in superalloys produced by powder metallurgy. The behavior of short cracks nucleated from artificial defects was investigated under small scale yielding. A deterministic prediction of the life of smooth specimens was made using the growth law measured in air or in vacuum. The distribution of real defects was then used in a probabilistic life prediction model. The growth of small cracks under large scale yielding was investigated in stainless steels. A modified Tomkins equation can account for the behavior of small cracks and provide an estimate of the lifetime under low-cycle fatigue or thermal fatigue. A damage model based on the propagation of micro-cracks originating at casting defects has been developed for single crystal turbine blades, operating under thermo-mechanical creep–fatigue conditions. The model used the process zone concept introduced by Mc Clintock. Weakening of material due to localized oxidation embrittlement is shown to account for oxidation–creep–fatigue interactions. The model gives satisfactory life predictions under various thermo-mechanical loading conditions. A local approach to fracture is proposed for fatigue crack growth using a fracture criterion as a post-processor of a finite element model, for two-dimensional long cracks.