In order to explore the improving effect and mechanisms of vegetation communities on soil structure and infiltration processes, we used fractal scaling theory to analyse soil particle-size distribution (PSD), soil dimension and soil infiltration for seven vegetation communities in the Dabie mountainous region of central China. The results showed that coniferous and broad-leaved mixed forests (PQ) had a higher function of meliorating soil particle structure and infiltration capability than broad-leaved forests (PC and QA) and coniferous forests (PD). In general, the amounts of silt and clay increased under PQ, PC and QA vegetation communities, whereas fine sand content decreased, resulting in higher values for soil total porosity and capillary porosity. For shelter forests, the infiltration rate was higher than other vegetation communities. The overall fractal dimensions of PSD ranged from 2.071 to 2.430, and the fractal dimensions of PQ, PC, QA and PD in shelter forests were far higher (mean value of 2.312) than those of the others. There was significant positive correlation between the fractal dimension and the amount of silt and clay (R = 0.815), and negative correlation with fine sand (R = -0.549). There was also a strong linear positive relationship between the fractal dimension and the soil’s infiltration rate. Correlations of the fractal dimension with the soil infiltration capability provided strong evidence that vegetation communities enhanced the soil fractal dimension by increasing the amounts of silt and clay, thereby improving both aggregate structure and pore structure and enhancing the degree of soil texture uniformity and infiltration capacity. This study demonstrates that fractal dimension analysis may be used to quantify differences in PSD and soil infiltration capability better; furthermore, the results can provide a reference for scientific selection and the distribution of plant types that will support soil and water conservation.