This paper presents an investigation of suitable failure criteria for predicting the strengths of uncracked and interfacially-cracked adhesively-bonded joints. A detailed experimental study of both bulk adhesive and adhesive joint behaviour has been carried out. The effect of both strain rate and temperature on the response of the adhesive to mechanical loading has been investigated through a series of tensile tests. The resulting data were used to construct an empirical model for the behaviour of the adhesive. A novel test method based on a four-point bend specimen has been used to investigate how the hydrostatic stress affects the response of the adhesive. Extensive tests on adhesive joints, subjected to different modes of loading and different lengths of interfacial cracks, have provided comprehensive joint strength data and insight into the site and locus of joint failure initiation. Following this, various failure criteria have been evaluated by carrying out detailed linear elastic and non-linear elasto-plastic two-dimensional analyses of the joints tested. Three-dimensional analyses provided modified loads for these two-dimensional analyses that more accurately reproduce the conditions on the plane of failure. Criteria based on critical stress or strain components at a distance from the point of singularity were investigated. A procedure for accounting for the strain rate effects of the adhesive has been incorporated with the non-linear analyses. Criteria based on critical energy release rates have been evaluated from the linear elastic analyses of the joints with interfacial cracks diminishing to very small sizes. Finally, non-linear springs along a plane of failure have been used to model a line of localised damage, resulting in joint failure criteria based on a critical opening displacement. This last method provides the most physically acceptable way of predicting the strength of cracked and non-cracked joints using the same failure criterion.