Fatigue crack initiation and propagation were investigated in structural adhesive joints consisting of 7075T6 aluminium adherends bonded with a mineral filled structural epoxy (Cybond 4523GB, American Cyanamid). Three types of joints were tested to achieve mode I (double-cantilever beam specimen, DCB), mixed mode I–II (cracked lap shear specimen, CLS), and mode II (end notch flexure specimen, ENF). All tests were conducted under ambient conditions with load ratio of 0.1 at a frequency of 30 Hz. Fatigue loading significantly reduced the strain energy release rate (G) required to initiate a crack compared with static and quasi-static loading. For the load ranges tested, fatigue precracks doubled the time to cause a resumption of crack growth under mode I loading. Negligible differences in crack initiation times (time to generate a crack from a fillet or resume extension of an existing crack) were observed for mixed-mode I–II and mode II specimens with cracks starting from fast mode I precracks, intact fillets and fatigue precracks. For the adhesive system tested, the relative influence of the mode ratio depended on whether the rate of crack propagation was plotted versus G max or % G c (percentage of the quasi-static critical energy release rate at the particular mode ratio). When expressed as a function of % G c, debonding rates were greatest under mixed-mode conditions at a given % G c, and were indistinguishable under mode I and mode II loading. However, when expressed as a function of G max, the propagation rates at a given G max were the same under mixed-mode and mode I loading, and smaller under mode II loading. This means that the allowable loads for joints in fatigue will depend on the mode ratio; for mixed-mode joints it will be a smaller fraction of the quasi-static allowable load than for mode I or mode II joints. Threshold energy release rates ( G max) under mode I and mixed mode I–II loading were essentially the same, and were obtained equally from extrapolated crack propagation rates or crack initiation times. For this adhesive system, it is recommended that adhesive joint design be based on threshold values for zero crack growth, because crack propagation rates show too much scatter to be relied upon for the prediction of in-service subcritical crack growth, particularly under mode I and mode II loading.