Sediment fingerprinting technology is widely used to differentiate sediment sources. However, despite its long-recognized benefits, there it has been seldom applied to assess the variability of sediment sources during storm events. In this study, sediment fingerprinting is used for four storm events to determine the dynamic changes in sediment sources throughout them in the black soil region in Northeast China. Three potential sediment sources—cultivated land, unpaved roads, and gullies—were effectively differentiated using four geochemical tracers (As, Be, Cs, and Cu), with an accuracy of 100%. The relative sediment contribution from each source was determined using linear and Bayesian mixing models. The mean absolute fit (MAF) values of the linear mixing model (MAFmean = 0.976–0.949) were higher than those of the Bayesian mixing model (MAFmean = 0.921–0.992), indicating that the first performed better. Cultivated land was the primary source of the sediment load, accounting for 59.03% of it (load-weighted mean = 68.29%), followed by the gullies (37.15%, load-weighted mean = 28.09%), and unpaved roads (3.90%, load-weighted mean = 3.69%) for the four storm events. In addition, a high variability in sediment source contribution was observed during the storm events. Cultivated land was the dominant sediment source during storm events with higher sediment concentrations (logarithmic function, r2 = 0.878, p < 0.01), discharge (linear function, r2 = 0.452, p < 0.05), and sediment flux (logarithmic function, r2 = 0.857, p < 0.01), whereas the reverse was observed for gullies. Contrastingly, the contribution of sediment from unpaved roads remained relatively stable during rainfall events. This provides a potential means to assess dynamic changes in sediment contributions from different erosion units. Moreover, it provides data support for exploring soil erosion mechanisms and effective erosion control in the black soil region in Northeast China.