In the traditional approach to fracture mechanics, the stress analysis is separated from a description of the actual process of material failure. A parameter characterizing the crack tip field, e.g. the energy release rate, is assumed to be a material property and known from experiment. Crack initiation is identified with this parameter reaching a critical value and the description of continued crack growth depends on knowing this parameter as a function of the amount of crack growth and the crack speed. Here, an alternative approach is described where the failure characteristics are embodied in a phenomenological constitutive relation that describes separation along one or more cohesive surfaces. Constitutive relations are specified independently for the material and for the cohesive surfaces. Fracture emerges as a natural outcome of the deformation process, without introducing an additional failure criterion. The characterization of the mechanical response of a cohesive surface involves both an interfacial strength and the work of separation per unit area, which introduces a characteristic length into the formulation. This framework has been used to address issues including void nucleation, quasi-static crack growth, dynamic crack growth, thermal crack growth and reinforcement cracking in metal-matrix composites.