Abstract The variation of the grain boundary self-diffusion coefficient with applied stress and the production of vacancies in the lattice are applied to the case of cavity growth in high temperature fatigue with particular reference to the case of magnesium at 400°c. It is shown that cavity growth should increase with increasing testing frequency and that there is a critical condition for growth to take place. An outline is given of a theoretical estimate of times to fracture and it is shown that the time of testing, rather than the total number of cycles, becomes the important endurance parameter. Predictions of the model are compared with experimental findings.