The main objective of this study was the source attribution of the water-soluble oxidative potential of PM2.5 in Xiamen, China. Mean concentrations of individual water-soluble trace elements varied within a wide range from <1 ng m−3 to >100 ng m−3 between November 2015 and May 2018. Their fractional solubility ranged from <5% to >70%, depending mainly on the origin, followed by meteorological conditions and atmospheric aging. The annual mean value of the volume-normalized dithiothreitol activity (DTTv) in PM2.5 between January 2017 and May 2018 in the urban area (0.632 ± 0.299 nmol min−1 m−3) was higher than in the suburban area (0.562 ± 0.247 nmol min−1 m−3) (p < 0.05), which was consistent with the spatial difference of transition metals and carbon species. DTTv were lower in summer than in winter (p < 0.05), consistent with the variations in Cu and Mn, which accounted for more than 94% of the reconstructed DTTv. DTTv could not be predicted well by lifetime cancer risk (LCR) and hazard index (HI), because the main contributors to HI (V, Ni, and Mn) and LCR (Cr(VI) and V) were different from those of DTTv. Based on the positive matrix factorization (PMF) and the multiple linear regression (MLR) models, the identified sources of DTTv at the urban/suburban sites were as follows: vehicle exhaust + biomass burning (39.9%/24.2%), primary organic aerosol (19.5%/23.9%), secondary nitrate/chlorine (10.5%/18.4%), secondary sulfate + ship emissions (18.2%/7.8%), mineral dust (2.3%/20.0%), and coal combustion (9.7%/5.6%). The results differed from those for PM2.5, LCR and HI, which had larger contributions from coal combustion and secondary sulfate + ship emission.