Objective:Identifying individuals at the earliest stages of Alzheimer’s Disease (AD) would enable development and study of interventions prior to onset of symptoms. However, differentiating age-related cognitive changes from subtle pathological changes remains a challenge in the field. Methods that would enable earlier detection of AD in elders with no subjective or objective cognitive concerns (i.e., individuals in the preclinical stage) would be of great interest. Community detection, a metric founded in graph theory, offers an alternative approach for characterizing subtle heterogeneity within aging samples and has the potential to inform cognitive variability in aging.Participants and Methods:Using a hierarchical community detection, we examined whether cognitive subtypes could be identified in 226 cognitively normal older adults (from the Alzheimer’s Disease Neuroimaging Initiative [ADNI] study). Cognitive profiles of each community were characterized first using MANOVAs to examine the relationship between community membership and 12 age-, gender-, and education-corrected neuropsychological variables. Pair-wise comparisons were examined for significant main effects. We then examined whether these subtypes were related to biomarkers (cortical volumes, fluorodeoxyglucose (FDG)-positron emission tomography (PET) hypometabolism) or clinical progression. All p values were corrected for multiple comparisons.Results:Three communities (i.e., cognitive subtypes) were identified within the healthy aging sample. The first and largest community identified (N = 106) was characterized by a relative weakness on a single measure visuospatial executive function. Both the second (N = 76) and third community (N = 44) scored significantly lower on immediate, delayed, and recognition memory relative to the first community. The third community was characterized by a relative weakness in category fluency and speeded visual sequencing as well (p < .000). The three communities did not differ on age, gender, education, race, or ethnicity. Community membership was associated with entorhinal volume (with the second and third communities having significantly smaller entorhinal volumes than the first community), though community membership was not significantly associated with other biomarkers examined. Conversion rate reached trend level significance at 12 month follow up (more converters in the third community).Conclusions:Hierarchical community detection is an alternative method for characterizing neuropsychological variation and it appears sensitive to relatively small differences that may be observed in a normal aging sample. While the sample size was relatively small, this approach shows promise for potentially leading to earlier detection of cognitive decline among individuals classified to be aging normally (e.g., community 3).