The antibiotic distributions, partitioning, and migration pathways in river basins have withdrawn great attention in the past decades. This study investigates 26 antibiotics of five classifications in surface water and sediment samples at 23 sites in Fenhe River, a typical tributary of Yellow River. There are 21 antibiotics found in the water samples at the concentration from 113.8 to 1106.0 ng/L, in the decreasing order of SAs > QNs > MLs > TCs > CAs. Fifteen antibiotics were detected in the sediment at the concentrations from 25.11 to 73.22 μg/kg following the decreasing order of SAs > MLs > TCs > QNs > CAs. The antibiotic concentrations vary greatly in surface water, generally lower in upstream and in reservoirs, and reaching highest in the midstream of the Fenhe River after passing Taiyuan and Jinzhong, and then lower again in the downstream. The antibiotic concentrations in sediment have a less variation in the entire river basin, but become high in the downstream. The results show the water-sediment partitioning coefficients of antibiotics generally were lower than those in other areas, having a migration path from the water to suspended solids, and then accumulated in sediment. The water-sediment partitioning coefficients also vary across the basin. The water-sediment partitioning coefficients of sulfacetamide and tetracycline are higher than the water-sediment partitioning coefficients of other antibiotics, with less variation across the basin, the water-sediment partitioning coefficients of azithromycin, enrofloxacin, and roxithromycin are low in the midstream of the river, and high at the river source and downstream. The water-sediment partitioning coefficients are significantly affected by the pH of sediment and the particle size of sediment. The prediction models of water-sediment partitioning coefficients for antibiotics are constructed with the selected effecting factors. The simulation values of antibiotics except chlortetracycline and erythromycin are highly consistent with the observed values, indicating that the prediction model is reliable.