Introduction: The hypothalamic-pituitary-adrenal axis function in depression has been the subject of considerable interest, and its function has been tested with a variety of methods. We investigated associations between saliva cortisol at awakening and the 24-h urine cortisol output, both measured at study baseline, with endpoint depression scores. Methods: Patients were admitted to a psychiatric inpatient ward with a major depressive episode and were started on fixed duloxetine treatment. They delivered saliva samples at awakening and 15, 30, and 60 min post-awakening and sampled urine for 24 h. Subsequently, they started a daily exercise program maintained for a 9-week period. Clinician-rated depression severity was blindly assessed with the Hamilton Depression Rating 6-item subscale (HAM-D<sub>6</sub>). The cortisol awakening response was quantified by the area under the curve with respect to the ground (AUC<sub>G</sub>) and with respect to the rise (AUC<sub>I</sub>) using saliva cortisol levels in the 1-h period after awakening. Analysis of expected associations between depression severity, AUC<sub>G</sub>, AUC<sub>I</sub>, exercise, and 24-h cortisol output was performed in a general linear model. Results: In all, 35 participants delivered saliva or 24-h urine samples. The mean age was 49.0 years (SD = 11.0) with 48.6% females with a mean baseline HAM-D<sub>6</sub> score of 12.2 (SD = 2.3). In a statistical model investigating the association between HAM-D<sub>6</sub> at week 9 as a dependent variable and AUC<sub>I</sub>, concurrent HAM-D<sub>6</sub>, gender, smoking, and exercise volume as covariates, we found a significant effect of AUC<sub>I</sub>, concurrent HAM-D<sub>6</sub>, and exercise. The following statistics were found: AUC<sub>I</sub> (regression coefficient 0.008; F value = 9.1; p = 0.007), concurrent HAM-D<sub>6</sub> (regression coefficient 0.70; F value = 8.0; p = 0.01), and exercise (regression coefficient −0.005; F value = 5.7; p = 0.03). The model had an R<sup>2</sup> of 0.43. The association between HAM-D<sub>6</sub> endpoint scores and the AUC<sub>I</sub> showed that higher AUC<sub>I</sub> values predicted higher HAM-D<sub>6</sub> endpoint values. The association between HAM-D<sub>6</sub> endpoint scores and the exercise level showed that a high exercise level was associated with lower HAM-D<sub>6</sub> endpoint values. Conclusion: The results thus showed that high AUC<sub>I</sub> values predicted less improvement of depression and high exercise levels predicted more improvement of depression. These findings need to be confirmed in larger samples to test if more covariates can improve prediction of depression severity.