Objectives: Studies of the demography of past populations involving deterministic life tables can be criticized for ignoring the errors of estimation. Bayesian methods offer an alternative, by focusing on the uncertainty of the estimates, although their results are often sensitive to the choice of prior distributions. The aim of this study is to explore a range of Bayesian methods for estimating age at death for a population of nomadic warriors—Scythians from the Black Sea region. Materials and methods: In total, skeletons of 312 individuals (93 children and 219 adults) from Glinoe (Moldova), dated to the 5th–2nd century BCE, were examined. We unified the age categories corresponding to different aging methods, allowing an application of a probabilistic assessment of the age categorization. A hierarchical Bayesian multinomial-Dirichlet-Dirichlet model was applied, with a hypothetical, subjective reference population, a real reference population, and no reference. Results: Stationary-population life expectancy was estimated as 27.7 years (95% CI: 25.1–30.3) for a newborn (e₀), and 16.4 years (14.0–19.0) for 20-year-olds (e₂₀), although with high uncertainty, and sensitive to the model specification. Slight differences in longevity between different social strata and between the Classical and Late chronological periods were found, although with high estimation errors. A more robust finding, confirming earlier studies, was a high probability of death in young adulthood, which could depend on Scythian lifestyle (conflicts, wars). Discussion: Our study shows a way to overcome some limitations of broad age categorization by using the Bayesian approach with alternative model specifications, allowing to assess the impact of reference populations.