<bold>Objectives: </bold>Variability in various biomarkers has emerged as a new clinical indicator for diseases including neurodegenerative disorders. Gamma-glutamyl transferase (GGT) has a potential to be involved in the pathogenesis of dementia due to its function as a marker of oxidative stress and atherosclerosis. We investigated the association between baseline GGT, GGT variability, and dementia risk for the first time in a large population.<bold>Methods: </bold>The Korean National Health Insurance Service datasets of claims and preventive health check-ups from 2004 to 2016 were used for this retrospective longitudinal study. The risk of incident dementia (all-cause dementia, Alzheimer's disease, and vascular dementia) was analyzed according to sex-specific quartiles of baseline GGT and GGT variability, and groups categorized by baseline GGT and GGT variability in ≥40-year-old individuals without baseline dementia (N = 6 046 442; mean follow-up 6.32 years).<bold>Results: </bold>During follow-up, 166 851 cases of new dementia developed. The fully adjusted hazard ratios (HRs) and 95% confidence intervals (CIs) for incident dementia increased in the higher quartiles of baseline GGT and GGT variability (HR [95% CI]: Q2, 1.034 [1.019-1.049]; Q3, 1.090 [1.075-1.105]; Q4, 1.212 [1.196-1.229]). The association between GGT variability quartiles and dementia risk remained significant even after adjusting for log-transformed baseline GGT level. The fully adjusted HRs for dementia was highest in the group with high baseline GGT concentration and the highest GGT variability quartile [HR (95% CI): 1.273 (1.250-1.296)].<bold>Conclusions: </bold>Not only baseline GGT level, but also GGT variability may be an independent predictor of dementia, and might be used for risk stratification for future dementia.