Abstract: Managing abandoned cropland requires knowledge of its extent and spatial distribution, as well as an assessment of the likelihood that individual abandoned parcels will return to cultivation. Existing studies of recultivation potential mainly consider terrain, soil, location and related conditions. They give less attention to differences in actual land use outcomes reflected in historical recultivation behavior. This study therefore assessed the recultivation potential of existing abandoned cropland in Sichuan Province, China, using evidence from historical recultivation behavior. This study first constructed an annual land cover dataset for Sichuan Province at a spatial resolution of 30 m for 1997–2025 using Landsat imagery. This dataset provided the basis for identifying cropland abandonment and recultivation each year and analyzing their spatial and temporal changes. Historical abandonment–recultivation events were then used as samples to develop a Stacking model for predicting the probability of recultivation of abandoned cropland. Shapley additive explanations (SHAP) analysis quantified the contributions of individual factors to model predictions and identified the importance and direction of the contributions of the main factors. Finally, the model was applied to cropland that remained abandoned in Sichuan Province in 2025 to assess its recultivation potential. The spatial distributions of abandonment extent and recultivation potential were also compared at the county level. The results showed that: (1) From 2000 to 2025, the area of remaining abandoned cropland in Sichuan Province increased continuously from 1,897.48 km² to 35,034.20 km². Annual recultivated area fluctuated during the same period. Changes in the area remaining abandoned and in annual recultivated area were not synchronized. (2) The number of consecutive years of abandonment was the most important predictor of recultivation probability, followed by agricultural output per unit area, mechanization, cropland connectivity, elevation, and slope. SHAP analysis showed that the contribution of consecutive years of abandonment to model predictions shifted from positive to negative after approximately seven years. Slope generally made a negative contribution to model predictions when it exceeded approximately 20°. (3) In 2025, low, moderate, and high recultivation potential zones accounted for 98.23%, 1.19%, and 0.58% of the abandoned cropland area, respectively. Low recultivation potential characterized the vast majority of the area assessed. The county-level comparison showed that the spatial distribution of abandonment extent did not fully correspond to that of recultivation potential. This study incorporated differences in actual land use outcomes reflected in historical recultivation behavior to complement theoretically based assessments of recultivation potential. The results provided a reference for identifying abandoned cropland for priority verification, classifying recultivation potential, and developing differentiated management strategies.