Objective: Melatonin has been considered to have an essential role in the pathophysiology of Alzheimer’s disease (AD) for its regulatory function on circadian rhythm and interaction with glutamate for the modulation of learning and memory. Previous studies revealed that melatonin levels decreased in patients with AD. However, melatonin supplement didn’t show promising efficacy for AD. This study compared trough melatonin levels among elderly people with different severities of cognitive deficits. Methods: We enrolled 270 elder individuals (consisting four groups: healthy elderly, amnestic mild cognitive impairment [MCI], mild AD, and moderate-severe AD) in the learning cohort. Trough melatonin levels in plasma were measured using ELISA. Cognitive function was evaluated by Clinical Dementia Rating Scale (CDR) and Mini-Mental State Examination (MMSE). An independent testing cohort, also consisting of four groups, was enrolled for ascertainment. Results: In the learning cohort, trough melatonin levels decreased in the MCI group but elevated in the mild and moderate to severe AD groups. Trough melatonin levels were associated with CDR and MMSE in MCI or AD patients significantly. In the testing cohort, the results were similar to those in the learning cohort. Conclusion This study demonstrated that trough melatonin levels in the peripheral blood were decreased in MCI but increased with the severity of AD. The finding supports the trials indicating that melatonin showed efficacy only in MCI but not in AD. Whether trough melatonin level has potential to be a treatment response biomarker for AD, especially its early phase needs further studies.