Oxidative stress is a central mechanism underlying the toxicity of fine particulate matter (PM_2.5); however, the source-specific chemical drivers of particle-associated oxidative potential remain incompletely understood. In this study, the oxidative potential (OP) of ambient PM_2.5 in Xi’an was investigated during winter and summer using the dithiothreitol (DTT) assay, with particular emphasis on the toxicological roles of water-soluble metals and emission sources. PM_2.5 exhibited significantly higher volume-normalized OP (DTT_v) in winter, indicating an enhanced particle-associated oxidative stress burden during the heating period. Notably, although water-soluble metals accounted for only a minor fraction of PM_2.5 mass, interpretable machine learning analysis (XGBoost–SHAP) identified potassium and manganese as dominant contributors to OP, highlighting the importance of biomass burning tracers and redox-active transition metals in particle-mediated reactive oxygen species generation. Source apportionment further revealed pronounced seasonal contrasts: dust sources contributed substantially to wintertime OP primarily due to their large mass loading, whereas traffic-related emissions dominated OP in summer owing to their high intrinsic oxidative toxicity. Overall, these findings suggest that variations in PM_2.5 oxidative potential are more closely associated with chemical composition and source-specific oxidative activity than with particle mass alone, providing additional insight into the factors influencing PM-related health risks.