Background Results on the association between depression and the hypothalamo-pituitary-adrenal (HPA) axis have been inconsistent, possibly due to heterogeneity of the DSM-IV category of depression. Specific symptom-dimensions could be used as a more homogenous phenotype in HPA-axis research. Methods Subjects (n = 1029) with a lifetime depression and/or anxiety disorder from the NESDA study (Netherlands Study of Depression and Anxiety) (mean age: 43.0 ± 12.7 years, 67.4% women) provided seven saliva samples to yield the cortisol awakening response (CAR), evening cortisol, and dexamethasone suppression data. The dimensions of the tripartite model (General Distress, Anhedonic Depression, and Anxious Arousal) were measured with the 30-item adapted Mood and Anxiety Symptoms Questionnaire (MASQ-D30) and analyzed in association with the cortisol measures with linear and nonlinear regression. Results Median (interquartile range) scores of General Distress, Anhedonic Depression, and Anxious Arousal were 20 (1427), 36 (2844), and 15 (1219), respectively, indicating large variability. Nonlinear associations with the shape of an inverted U were found between General Distress, Anhedonic Depression, and Anxious Arousal on one hand and total morning secretion and the dynamic of the CAR by contrast. Both high and low severity levels were associated with a lower CAR, compared with intermediate levels of severity. Most of the associations remained significant when adjusted for covariates and the presence of DSM-IV diagnoses. Conclusions Nonlinear associations were found between the CAR and the dimensions of the tripartite model. This could explain previous inconsistent findings regarding HPA-axis activity in depressed patients and illustrates the added value of symptom-dimensions for HPA-axis research. © 2011 Society of Biological Psychiatry.