The flow–sediment relationship is important to understand soil erosion and sediment transport in severely eroded areas, such as Loess Plateau. Previous research focused on the variation and driving forces of runoff and sediment at the different scales in a watershed. However, the variations of the flow–sediment relationship on multispatial scales (slope, subgully, gully, and watershed scales) and multitemporal scales (annual, flood events, and flood process) were less focused. Taking the Peijiamao watershed, which includes whole slope runoff plot (0.25 ha, slope scale), branch gully (6.9 ha, subgully scale), gully (45 ha, gully scale), and watershed (3930 ha, watershed scale), four different geomorphic units located at the Chinese Loess Plateau, as the research site, a total of 31 flood events from 1986 to 2008 were investigated, and two flood process data were recorded across all the four geomorphic units. The results showed that on the annual timescale, the average sediment transport modulus and runoff depth at four scales exhibited a linear relationship, with determination coefficients of 0.81, 0.72, 0.74, and 0.77, respectively. At the flood event timescale, the relationships between sediment transport modulus and runoff depth at the gully and watershed scales could also be fitted with a linear relationship with high determination coefficients (from 0.77 to 0.99), but the determination coefficient at the slope scale was only 0.37 at the event scale. On the single rainfall event timescale, the flow–sediment relationship at the slope scale showed a figure-eight hysteretic pattern while those relationships at larger scales showed an anticlockwise loop hysteretic pattern. Under the same flow condition, the suspended sediment concentrations during the falling stage were significantly higher than those during the rising stage. Moreover, the difference was bigger as the spatial scale increased due to the wash loads in the downstream gullies, which favored the occurrence of hyper-concentration flow. The results of the study could provide useful insights into the temporal–spatial scale effects of sediment transport and their internal driving mechanisms at the watershed scale.