Abstract: Soil aggregates can store and supply the organic matter in soil. The structure and stability of the soil aggregates are two of the most key indicators in cultivated fields. The soil structure and erosion resistance are then evaluated for the ecological restoration. However, it is still lacking in the underlying mechanisms of the stability of the soil aggregate under long-term ecological restoration. This study aims to identify the effects of the Grain-for-Green project (GFGP) on the composition and stability of the soil aggregates. Four sample plots were selected from the tillage layer of the calcareous purple soil region in the central Sichuan basin, China. The GFGP was then implemented with different years, including the cultivated land (CL), Land with 25 years of GFGP (GFGP25), Land with 25 years of GFGP (GFGP50), and forest land (FL). A survey of the layered profile was conducted on the spatial distribution of the soil aggregate composition in the 30 cm tillage layer of the different lands. The stability of the soil aggregate was characterized by some indicators, such as the percentage proportion of the dry-sieved aggregates larger than 0.25 mm (DR0.25), the percentage proportion of water-stable aggregates larger than 0.25 mm (WR0.25), mean weight diameter (MWD), geometric mean diameter (GMD), percentage of aggregate disruption (PAD), water stable aggregates ratio (WSAR), soil erodibility factor (K) and Fractal Dimension (D). Moreover, the partial least squares path model (PLS-PM) was used to explore the key driving factors of the GFGP on the soil aggregate stability. The results revealed that the trends of the uniform distribution were observed in the tillage profiles for the different grain sizes of the mechanically and water-stable aggregates in the CL. Moreover, the profile distributions of the soil aggregates in the GFGP25, GFGP50, and FL were primarily different in the content of the mechanically stable aggregates of >5 mm and water stable aggregates of >2 mm, thus fluctuating, increasing, and decreasing trends along the vertical direction, respectively. Notably, there was the maximum content of the mechanically stable aggregates of >5 mm in each land at the mid-slope location. While there was a smaller content of the water stable aggregates of >2 mm. Soil aggregate stability shared a relatively uniform distribution in the tillage layer (0–30 cm) within the CL. By contrast, the soil aggregate stability was significantly higher in the surface layer, compared with the subsurface layer in the GFGP25, GFGP50, and FL. Furthermore, the soil aggregate stability on the slope displayed a relatively uniform distribution in the GFGP25. While at the mid-slope position, there was significantly lower soil aggregate stability compared with the rest locations in the CL, GFGP50, and FL, which was primarily attributed to the erosion process. Compared with the CL, the GFGP significantly enhanced the soil aggregate stability in the tillage layer, with the increase in the WR0.25, WSAR, MWD, and GMD ranging from 33.37% to 41.59%, 33.96 to 44.64%, 23.18 to 36.02%, and 49.02 to 83.77%, respectively. The soil physicochemical properties and soil aggregate composition were enhanced with an indirect effect coefficient as high as 0.717. Soil organic matter and bulk density were identified as the key driving factors. Overall, this finding can also provide a strong reference to improve the soil structure and quality in the future implementation and planning of the GFGP.