The increasing frequency and variability of extreme rainfall are escalating soil erosion risks globally. However, the synergistic effects of extreme rainfall and heterogeneous slope surface conditions on runoff and soil loss remain poorly quantified in Southwest China's karst regions. This study compiled a dataset of 1,718 runoff plot-years from 16 monitoring sites in Guizhou Province (2006-2022) spanning diverse land use types-bareland, cropland, grassland, shrubland, orchard, and forestland. The effects of extreme rainfall, land use/cover, and topographic factors on runoff and erosion on karst hillslopes were assessed, with their relative contributions quantified using multivariate regression and path analysis. Results show that: (1) Extreme rainfall, defined as precipitation (P) exceeding 57.05 mm or maximum 30-minute intensity (I-30) of 41.89 mm<middle dot>h(-1), amplifies runoff by 2.35- to 4.83-fold and soil erosion by 4.12- to 21.07-fold compared to normal rainfall, contributing 36.00-47.87 % of total runoff and 40.14-74.20 % of soil loss across land uses. (2) Karst slopes exhibited low annual runoff coefficients (<10 % for bare land, similar to 2 % for cropland, and similar to 1 % for vegetated lands). Erosion is most severe on bare land (1,733.93 t<middle dot>km(-2)<middle dot>yr(-)(1)), followed by cropland (286.13 t<middle dot>km(-)(2)<middle dot>yr(-)(1)), whereas vegetated lands (11.05-56.04 t<middle dot>km(-2)<middle dot>yr(-)(1)) was constrained to a tolerable level. (3) Erosion exhibited a non-linear response to slope gradient, with critical thresholds at 18.73 degrees for bareland and 32.56 degrees for vegetated slopes, beyond which erosion rates attenuated. (4) Runoff generation is chiefly regulated by rainfall depth (P), whereas soil erosion is dominated by runoff depth (RD) and 30-minute maximum intensity (I-30), with slope gradient (S), antecedent soil water content (SWC), and vegetation traits as key mediators. However, the complexity of surface and subsurface interactions in karst landscapes limits the model's explanatory power, capturing no more than 60 % of runoff and erosion variability. These findings provide critical insights for sustainable land management and policy interventions to mitigate soil erosion and combat rocky desertification in fragile karst ecosystems.