Study region: Qilian Mountain National Park (QLMNP) in northwestern China. Study focus: The risk of soil erosion has increased noticeably at the global scale. In China, several ecological projects have been conducted to minimize the side effect of soil erosion in recent decades. Assessing soil erosion and identifying driving mechanisms are necessary due to the increased risk of ecological security in QLMNP. Here, we employed the Revised Universal Soil Loss Equation (RUSLE) to assess the soil erosion rate in QLMNP from 1982 to 2020. The driving mechanism was detected through the differences between the realistic scenario and four idealized scenarios with different fixed RUSLE input parameters. The contributions of different driving factors (soil type, digital elevation model, slope, extreme precipitation frequency and intensity, precipitation, vegetation coverage, and land use/cover change (LUCC)) were assessed through the geodetector model. New hydrological insights for the region: The soil erosion rate averaged over QLMNP increased at 0.13 t ha-1 yr-1 from 1982 to 2020, despite the slightly decrease shown in the majority (62.62%) area. Anthropic LUCC in eco-conservation measures tends to prevent soil erosion in QLMNP, but extreme precipitation has the opposite effect. The coupled impacts of extreme precipitation and LUCC dominate the soil erosion dynamics in this area. Our results indicate that ecological projects are effective to mitigate the soil erosion caused by climate change. © 2022 The Authors