AbstractBackgroundPopulations are aging worldwide, average life expectancy in the U.S. has stagnated around 80 years old, and prevalence of Alzheimer’s disease and related dementias are increasing. The Hayflick limit predicts a maximum lifespan of about 120 years among humans, and the oldest observed living human, Jeanne Calment, died after her 122nd birthday. In other words, there appears to be a wide gap between actual and potential human lifespans. Furthermore, the onset age for most health conditions has not increased markedly in many decades. This suggest that increases in lifespan may stem from extensions in the duration of sickness prior to death. Preventing morbidity and increasing healthspan will require better understanding of the different ways individuals can age. Typically, brain aging is accompanied by volumetric changes that negatively impact most cognitive processes. Despite what happens to the majority, or what happens on average, there is a vast range of variation in neurocognitive development in late life. Charting the distinctiveness of developmental trajectories within a population will help identify meaningful subpopulations that could not be uncovered ex ante. In this study we investigated cognitive changes over time and developed typologies of different and unique ways people experienced brain aging.MethodFluid intelligence data among 17,016 UK Biobank subjects was sampled three times from 2006 until 2014. Growth curves were used to estimate everyone’s level and rate of change. With these trajectory values, a cluster analysis distinguished several cognitive aging trajectories.ResultWe identified up to seven distinguishable trajectories. Each trajectory component was significant at p<.001. Two trajectories exhibited learning and other cognitive improvements over time, three maintained their cognitive performance, and two experienced cognitive declines over the eight‐year period.ConclusionAmong longitudinal cohorts, cluster analysis applied to growth curves identified meaningful subpopulations without the need for expert diagnosticians. Including typological information in population models promises to reduce within‐group variance and allow testing for processes that are conditional rather than unanimous across a population. Future works should investigate how trajectories are related to one another and whether more diverse populations demonstrate similar trajectory typologies.