Understanding the sediment–water partitioning of emerging contaminants is essential for assessing their mobility, persistence and ecological risks in river systems. However, the mechanisms driving large differences in partitioning behaviour across contaminant classes remain poorly resolved at the basin scale. Here we show that major emerging contaminant classes follow distinct mechanistic regimes in sediment–water partitioning. Analysis of a nationwide dataset comprising 5085 paired sediment–water records across China's major river basins reveals that antibiotics are predominantly controlled by molecular descriptors, per- and polyfluoroalkyl substances are strongly modulated by ion-mediated interfacial processes, and endocrine-disrupting chemicals exhibit synergistic regulation by molecular, geochemical and basin-scale factors. Integration with molecular dynamics simulations confirms these class-specific mechanisms at the molecular interface, while a multi-branch multi-head attention framework enables accurate prediction (R² = 0.76–0.92) and spatially explicit mapping of high-accumulation versus high-mobility zones. By identifying class-specific mechanistic regimes, this work provides both improved predictive capability and a clearer mechanistic basis for basin-scale risk assessment of emerging contaminants.