Background: As Alzheimer’s disease (AD) pathology presents decades before dementia manifests, unbiased biomarker cut-points may more closely reflect presence of pathology than clinically defined cut-points. Currently, unbiased cerebrospinal fluid (CSF) tau cut-points are lacking. Methods: We investigated CSF t-tau and p-tau cut-points across the clinical spectrum using Gaussian mixture modelling, in two independent cohorts (Amsterdam Dementia Cohort and ADNI). Results: Individuals with normal cognition (NC) (total n = 1111), mild cognitive impairment (MCI) (total n = 1213) and Alzheimer’s disease dementia (AD) (total n = 1524) were included. In both cohorts, four CSF t- and p-tau distributions and three corresponding cut-points were identified. Increasingly high tau subgroups were characterized by steeper MMSE decline and higher progression risk to AD (cohort/platform-dependent HR, t-tau 1.9–21.3; p-tau 2.2–9.5). Limitations: The number of subjects in some subgroups and subanalyses was small, especially in the highest tau subgroup and in tau PET analyses. Conclusions: In two independent cohorts, t-tau and p-tau levels showed four subgroups. Increasingly high tau subgroups were associated with faster clinical decline, suggesting our approach may aid in more precise prognoses.