Introduction: Neuropsychiatric symptoms are important treatment targets in the management of dementia and can be present at very early clinical stages of neurodegenerative diseases. Increased cortisol has been reported in Alzheimer’s disease (AD) and has been associated with faster cognitive decline. Elevated cortisol output has been observed in relation to perceived stress, depression, and anxiety. Dehydroepiandrosterone sulfate (DHEAS) has known anti-glucocorticoid effects and may counter the effects of cortisol. Objectives: We aimed to examine whether CSF cortisol and DHEAS levels were associated with (1) neuropsychiatric symptoms at baseline, (2) changes in neuropsychiatric symptoms over 3 years, and (3) whether these associations were related to or independent of AD pathology. Methods: One hundred and eighteen participants on a prospective study in a memory clinic setting, including patients with cognitive impairment (n = 78), i.e., mild cognitive impairment or mild dementia, and volunteers with normal cognition (n = 40), were included. Neuropsychiatric symptoms were assessed using the Neuropsychiatric Inventory Questionnaire (NPI-Q). CSF cortisol and DHEAS, as well as CSF AD biomarkers, were obtained at baseline. Neuropsychiatric symptoms were re-assessed at follow-up visits 18 and 36 months from baseline. We constructed linear regression models to examine the links between baseline neuropsychiatric symptoms, the presence of AD pathology as indicated by CSF biomarkers, and CSF cortisol and DHEAS. We used repeated-measures mixed ANCOVA models to examine the associations between the neuropsychiatric symptoms’ changes over time, baseline CSF cortisol and DHEAS, and AD pathology. Results: Higher CSF cortisol was associated with higher NPI-Q severity scores at baseline after controlling for covariates including AD pathology status (B = 0.085 [0.027; 0.144], p = 0.027; r = 0.277). In particular, higher CSF cortisol was associated with higher baseline scores of depression/dysphoria, anxiety, and apathy/indifference. Elevated CSF cortisol was also associated with more marked increase in NPI-Q scores over time regardless of AD status (p = 0.036, η² = 0.207), but this association was no longer significant after controlling for BMI and the use of psychotropic medications. CSF DHEAS was associated neither with NPI-Q scores at baseline nor with their change over time. Cortisol did not mediate the association between baseline NPI-Q and changes in clinical dementia rating sum of boxes over 36 months. Conclusion: Higher CSF cortisol may reflect or contribute to more severe neuropsychiatric symptoms at baseline, as well as more pronounced worsening over 3 years, independently of the presence of AD pathology. Our findings also suggest that interventions targeting the HPA axis may be helpful to treat neuropsychiatric symptoms in patients with dementia.