Amongst a large number of fatigue criteria proposed for the prediction of crack initiation in thermomechanical fatigue, various approaches have been found to be particularly useful for certain categories of material over specific domains of temperature and cyclic strain. However, no particular approach appears to give invariably better predictions than others, so that the choice of the lifing model must be based on validation for the relevant circumstances.In this paper the focus is placed on the Energy Dissipation Criterion (EDC). We present physical arguments in favour of this approach's versatility, and illustrate its performance by the application of this approach both in the macroscopic and micromechanical context.Firstly, by way of illustration we consider cyclic deformation of a Ramberg-Osgood material, with a view to establish the equivalence between the EDC and some selected classical criteria. For this simple but analytically tractable case several interesting results can be established, including the equivalence between EDC and both stress range and strain range lifing criteria.Secondly, we consider fatigue loading of polycrystalline FCC material deforming by the combination of anisotropic linear elasticity and crystal slip. Energy dissipation density in this case is location-dependent even for a polycrystal subjected to macroscopically uniform stress and strain. Crack initiation then is predicted to occur at the 'weakest link' location corresponding to the most intense dissipation.The above two versions of energy dissipation criteria are each compared against experimental data. The comparative performance of Walker strain, Smith-Watson-Topper (SWT) and EDC lifing methods is assessed. It is concluded that EDC provides improved reliability, particularly for cases of complex loading paths and mechanisms interactions.