To assess soil quality in micro-topographic units of loess hilly regions, this study selected five typical micro-landforms (shallow gully, collapse, steep slope, gentle terrace, cut gully) along with the original slope as a control. We measured 23 indicators covering soil physicochemical properties, enzyme activities, and plant diversity. A soil quality evaluation model was constructed using principal component screening combined with fuzzy comprehensive evaluation, and structural equation modeling was applied to identify driving factors. The results showed that the first five principal components had a cumulative contribution rate of 79.915%. The SQI values ranked as follows: collapse (0.604) > gentle terrace (0.550) > cut gully (0.538) > shallow gully (0.530) > original slope (0.527) > steep slope (0.466), with the collapse micro-landform consistently ranking first across the physical, chemical, enzymatic, and plant diversity indicators. The structural equation models further showed that soil moisture had the strongest association with soil quality in collapses, shallow gullies, and the original slope, whereas soil organic matter showed the strongest association in steep slopes, gentle terraces, and cut gullies. These findings provide a scientific basis for precision soil quality management and ecological function enhancement based on micro-topographic units in the implementation of the Grain-for-Green program.