【Objective】 Soil health assessment is a critical technical approach for achieving sustainable farmland management. However, existing evaluation systems often suffer from limitations such as indicator redundancy and high operational costs, which hinder their widespread application. This study aims to construct a cost-effective and efficient minimum data set (MDS) for soil health evaluation in the semi-arid farmland regions of the Loess Plateau, and to scientifically validate its reliability and applicability under local ecological conditions. 【Method】A total of 100 soil samples were collected from dryland farmlands in Wuzhai County, Shanxi Province, a representative area of the Loess Plateau. A comprehensive set of 23 soil indicators covering physicochemical and biological properties was analyzed. The MDS was established through an integrated statistical procedure that combined the principal component analysis (PCA), norm value calculation, and Pearson correlation analysis to identify the most representative and non-redundant indicators. The soil health index (SHI) was subsequently calculated using both linear and nonlinear scoring functions based on the MDS and the total data set (TDS). The performance of the MDS was evaluated by comparing SHI values derived from both data sets and further validated through correlation analysis with crop yield data. 【Result】The MDS was successfully established and included six key indicators: soil bulk density, total nitrogen, urease, cellobiohydrolase, bacterial Shannon index, and fungal Shannon index. These indicators accounted for 82.47% of the total variance explained by the TDS. Notably, biological indicators constituted two-thirds of the MDS, underscoring the vital role of microbial processes in soil health within arid regions. The SHI values calculated using the MDS showed a strong and significant positive correlation with those from the TDS under both nonlinear and linear scoring functions (P < 0.001), confirming the MDS’s capability to effectively represent the full data set. Validation with crop yield data further demonstrated that the nonlinear scoring function applied to the MDS provided a better fit (r = 0.70) than the linear function (r = 0.64), indicating its superior suitability for soil health assessment in the regions. The average SHI across the studied area was 0.49, reflecting a moderate overall soil health status. Spatially, soil health exhibited a pattern of lower values in the north and higher values in the south, largely influenced by the high erodibility of loess soils and more pronounced aridity in the northern part. 【Conclusion】This study developed a simplified yet robust MDS for soil health evaluation in semi-arid farmland systems of the Loess Plateau, effectively balancing comprehensiveness and feasibility. The results highlight the essential role of microbial diversity and functional indicators, such as enzyme activities and bacterial/fungal diversity, in evaluating soil health under dryland conditions. The spatial variation in soil health calls for region-specific management strategies, particularly in northern areas where soil erosion and moisture limitation are more severe. It is recommended that future research place greater emphasis on incorporating microbial functional parameters into soil health assessment frameworks. Moreover, integrating emerging technologies such as soil sensing and molecular tools could further enhance the efficiency and predictive power of soil health monitoring in arid and semi-arid agricultural landscapes.