The failure of adhesively-bonded joints, consisting of metallic adherends and epoxy-based structural adhesive with a relatively low toughness ~200 J/m2, has been studied. The failure was via quasi-static mode I, steady-state crack propagation and has been modelled numerically. The model implements a ‘top-down approach’ to fracture using a dedicated steady-state, finite-element formulation. The damage mechanisms responsible for fracture are condensed onto a row of cohesive zone elements with zero thickness, and the responses of the bulk adhesive and of the adherends are represented by continuum elements spanning the full geometry of the joint. The material parameters employed in the model are first quantitatively identified for the particular epoxy adhesive of interest, and their validity is verified by comparison with experimental results. The model is then used to conduct a detailed study on the effects of (a) large variations in the geometrical configuration of the different types of specimens and (b) the adherend stiffness on the predicted value of the adhesive fracture energy, G a . These numerical modelling results reveal that the adhesive fracture energy is a strong nonlinear function of the thickness of the adhesive layer, the other variables being of secondary importance in influencing the value of G a providing the adhesive does not contribute significantly to the bending stiffness of the joint. These results which fully agree with experimental observations are explained in detail by identifying, and quantifying, the different sources of energy dissipation in the bulk adhesive contributing to the value of G a . These sources are the locked-in elastic energy, crack tip plasticity, reverse plastic loading and plastic shear deformation at the adhesive/adherend interface. Further, the magnitudes of these sources of energy dissipation are correlated to the degree of constraint at the crack tip, which is quantified by considering the opening angle of the cohesive zone at the crack tip.