OBJECTIVE To investigate the association between sleep-related characteristics and cognitive change over 3 years of follow up in an aged population. METHODS Sleep characteristics and covariates were assessed at baseline in a standardized interview and clinical examination of the population-based KORA Age Study (n = 740, mean age = 75 years). Cognitive score (determined by telephone interview for cognitive status, TICS-m) was recorded at baseline and 3 years later. RESULTS At baseline, 82.83% (n = 613) of participants had normal cognitive status, 13.51% (n = 100) were classified with mild cognitive impairment (MCI), and 3.64% (n = 27) with probable dementia. The effect of three distinct patterns of poor sleep (difficulties initiating [DIS] or maintaining sleep [DMS], daytime sleepiness [DS] or sleep duration) were considered on a change in cognitive score with adjustments for potential confounders in generalized linear regression models. Cognitive decline was more pronounced in individuals with DMS compared to those with no DMS (β = 1.33, 95% CI = 0.41-2.24, P < 0.001). However, the predictive power of DMS was only significant in individuals with normal cognition and not impaired subjects at baseline. Prolonged sleep duration increased the risk for cognitive decline in cognitively impaired elderly (β = 1.86, 95% CI = 0.15-3.57, P = 0.03). Other sleep characteristics (DIS and DS) were not significantly associated with cognitive decline. CONCLUSIONS DMS and long sleep duration were associated with cognitive decline in normal and cognitively impaired elderly, respectively. The identification of impaired sleep quality may offer intervention strategies to deter cognitive decline in the elderly with normal cognitive function.