As one of the largest and most typical continuous karst distribution belts in the world, Southwest China is vulnerable to soil erosion and rocky desertification because of its unique geological conditions and heterogeneous landscapes. Understanding the rainfall-runoff-sediment transport relationships is critical to effectively controlling soil erosion in karst watersheds. This study aims to evaluate the sediment yield variability at different temporal scales (daily, monthly/seasonal, and annual) from 2014 to 2019 and to further identify the rainfall and runoff variables most associated with sediment yield in the selected Mahuangtian (2.26 km2) and Muzhaihe (0.09 km2) karst watersheds. Since rainfall and runoff variables were highly codependent, the partial least squares regression (PLSR) model was used to quantify the relative importance of rainfall and runoff variables to sediment yield. The variable importance of the projection (VIP) scores>1 was considered to be the most influential predictor of the PLSR model. Results showed that sediment yield was greatly influenced by rainfall and runoff, which can explain 70% and 85% of the variation of sediment yield in the Mahuangtian and Muzhaihe watersheds, respectively. For rainfall variables, three-day antecedent precipitation (VIP = 1.41) and five-day antecedent precipitation (VIP = 1.28) were the main first-order factors controlling sediment yield in the Mahuangtian watershed. In contrast, the rainfall amount (VIP = 1.35) was the dominant factor influencing sediment yield in the Muzhaihe watershed. For runoff variables, runoff amount, runoff coefficient, and maximum discharge were the primary variables controlling the sediment yield in these two karst watersheds. By partially eliminating the co-dependency of the variables, the PLSR approach can conveniently determine the dominant factors controlling sediment yield in heterogeneous karst watersheds.