Major depressive disorder (MDD) is a global problem for which current pharmacotherapies are not completely effective. Hypothalamic-pituitary-adrenal (HPA) axis dysfunction has long been associated with MDD; however, the value of assessing cortisol as a biological benchmark of the pathophysiology or treatment of MDD is still debated. In this review, we critically evaluate the relationship between HPA axis dysfunction and cortisol level in relation to MDD subtype, stress, gender and treatment regime, as well as in rodent models. We find that an elevated cortisol response to stress is associated with acute and severe, but not mild or atypical, forms of MDD. Furthermore, the increased incidence of MDD in females is associated with greater cortisol response variability rather than higher baseline levels of cortisol. Despite almost all current MDD treatments influencing cortisol levels, we could find no convincing relationship between cortisol level and therapeutic response in either a clinical or preclinical setting. Thus, we argue that the absolute level of cortisol is unreliable for predicting the efficacy of antidepressant treatment. We propose that future preclinical models should reliably produce exaggerated HPA axis responses to acute or chronic stress a priori, which may, or may not, alter baseline cortisol levels, while also modelling the core symptoms of MDD that can be targeted for reversal. Combining genetic and environmental risk factors in such a model, together with the interrogation of the resultant molecular, cellular, and behavioral changes, promises a new mechanistic understanding of MDD and focused therapeutic strategies.