Large amounts of sediment due to severe soil erosion in Sang-kan River Basin have a significant impact on the ecological environment of the basin and the water quality of Kwanting Reservoir. The study of sediment sources apportionment is of great importance to the formulation of soil and water conservation measures and the development of ecological environment construction. Aiming to explore the main source of sediment and the evolution of erosion and sediment yield in Sang-kan River Basin, the composite fingerprinting identification technology was used in this study. A typical yudiba dam was selected in Gaoqiang Township, Yangyuan County, Zhangjiakou City, North China. Three types of potential sediment sources were identified, which were gully wall, croplands, and forest-grasslands. Nine types of soil physical and chemical properties were measured, including soil particle size, SOC, TN, radionuclide (137Cs) and low-frequency magnetic susceptibility (Xlfa, Xlfb and Xlfc correspond to particle sizes of 2 mm, 0.063 mm and 0.05 mm respectively). The results show that the 137Cs average content of sediment was not significantly different from that of gully wall soil, and very significantly lower than that of forest-grasslands and croplands soil (p<0.01), which indicated that the sediment was mainly from gully wall. The whole siltation process was divided into three stages: stage Ⅰ (2007—2010), stage Ⅱ (2011—2013) and stage Ⅲ (2014—2017), based on changes in SOC and TN concentrations in the deposition profiles. As the majority of 137Cs content of gully wall was below the detection limit, 137Cs was a good indicator for identifying gully wall as the main sediment source but it was difficult to be used for multiple sediment sources in the small basin. The Kruskal-Wallis H nonparametric test was used to screen out 7 fingerprint factors with significant difference (p<0.05) among the different sediment sources, and the multiple stepwise discriminant function analysis was performed under the premise of Wilks'' lambda minimization. The optimum combination of fingerprint factors was composed of TN, Xlfb and SOC, and the cumulative correct discriminant rate of gully wall, croplands and forest-grasslands reached 86.1%. During the study period, the average contribution rates of sediment sources in the small basin were ranked in the descending order of gully wall (82.68±8.20%) > croplands (15.36±8.46%) > forest-grasslands (1.96±0.33%), and the goodness of fit ranged from 83.58% to 90.43%. The sediment contribution ratios per unit area of sediment sources were ranked in the descending order of gully wall (5.52) > croplands (1.55) > forest-grasslands (0.03), in stage Ⅲ. In the evolution of erosion and sediment yield, gully wall has long been the primary sediment source; croplands were the secondary sediment source, and its sediment contribution rate increased significantly when extreme rainstorm occurred, which was sensitive to extreme rainstorm; the sediment contribution rates of forest-grasslands were stable at low levels, and much lower than the area proportion of forest-grasslands in the small basin. Overall, the 137Cs tracing technique is suitable for identifying the main sediment source in the small basin, and the composite fingerprinting technique is better at identifying multiple sediment sources. The erosive rainfall was the main factor of soil erosion in Sang-kan River Basin, and the gully wall collapse caused by gully erosion was the main reason of serious soil erosion in the basin.