Abstract: This study aims to systematically reveal the intrinsic relationship between internal deformation and surface displacement in progressive retrogressive landslides. The research team conducted large-scale geotechnical centrifuge model tests using the Qing Shi landslide in the Three Gorges Reservoir area as the geological background, systematically investigating the evolution mechanism of the progressive failure process of landslides. (1) The test results show that surface deformation is an external manifestation of the progressive instability of the landslide, with the fundamental cause lying in the evolution of the internal sliding surface. Initially, the slope experiences slight settlement. As the test progresses, surface tensile cracks gradually deepen and expand, while shear deformation on the sliding surface intensifies and eventually connects. Once the sliding surface is fully connected, the first-level sliding block begins to slide, pulling the subsequent blocks, demonstrating the characteristic progressive retrogressive nature of the landslide. (2) A new method for classifying landslide stages based on displacement-time curves and the soil’s overconsolidation ratio is proposed. The landslide deformation process is divided into four major stages and eight sub-stages, including the initial consolidation stage and the instability of first, second, and third-level sliding blocks. (3) By studying the evolution rate of the main cracks in each sliding block, it was found that the first-level block enters the accelerated deformation stage at 0.035D, the second-level block at a rate below 0.024D, and the third-level block at 0.06D, revealing key moments for different blocks to enter the accelerated deformation stage.This study innovatively explores the internal and external deformation mechanisms of progressive retrogressive landslides through centrifuge simulations, proposes a new method for landslide stage classification, and offers an early warning index based on crack evolution rates. These findings are significant for improving landslide monitoring accuracy, optimizing prevention measures, and providing strong support for future landslide control and early warning systems, especially in complex geological conditions.