The contribution of telomere shortening to the onset of certain age-related diseases, such as dementia, and its role as a predictor of cognitive impairment remain unclear. We tested these hypotheses by analyzing telomere length in 449 inpatients in a large cohort of the oldest old (mean age 85 years) followed up yearly. No significant difference in telomere length was observed between cognitively normal patients (205), demented patients (195; 82 mixed dementia, 77 Alzheimer's disease and 21 vascular dementia) and patients (49) with mild cognitive impairment (MCI). Similarly, no significant differences in telomere length were found between patients with different etiologies or severities of dementia. Telomere length and change in cognitive status (from normal to MCI or dementia, or from MCI to dementia) were not associated after two years of follow-up. This longitudinal study in very old patients provided no evidence to suggest that telomere length could be used to distinguish between demented and non demented patients, regardless of the type of dementia, or to predict dementia or MCI conversion.