Soil microbial community shifts are sensitive indicators of ecological transitions in arid lands, but their use for tracing late-Holocene environmental variability in desert dune soils is still not well explored. Here, we investigated two desert dune systems in northwestern China (Mingsha and Kumtag) as model arid soil environments. These dune fields are remote and sparsely populated, with limited direct human disturbance. Stratified sampling across dune profiles provides a framework to compare microbial communities across layered edaphic contexts. We combined published paleoenvironmental background records, high-throughput sequencing of bacterial and archaeal communities, and community assembly modelling to examine stratified microbiome patterns across three chronological units, with particular attention to a focal interval (similar to 1.0-1.5 kyr BP). Approximate age assignments were compiled from published radiocarbon and optically stimulated luminescence constraints to provide temporal context. Across units, we observed distinct microbial assemblages. In the Mingsha Desert, bacterial communities shifted from methylotrophic taxa in the youngest unit to increased representation of spore-forming taxa in older strata, accompanied by archaeal shifts towards halophilic lineages. In the Kumtag Desert, compositional differences were comparatively weaker, consistent with strong local edaphic constraints. Quantitative assembly models indicated substantial contributions of stochastic processes (drift and dispersal limitation) across units, while the relative importance of deterministic filtering increased in strata characterized by higher salinity and stronger resource limitation. Co-occurrence network analyses further showed unit-dependent network structure and taxa occupying central network positions, providing candidate indicators of community reorganization across stratified edaphic contexts. Together, these results highlight the value of integrating regional paleoclimate context with stratified soil microbial data to characterize microbiome variation in dryland dune environments and to inform microbiome-based indicators relevant to desertification risk assessment.