Visible and Near-Infrared (Vis–NIR) spectroscopy has emerged as a rapid, non-destructive, and cost-effective technique for estimating soil properties and supporting digital soil mapping. This review critically evaluates the performance and limitations of Vis–NIR spectroscopy for predicting soil properties in arid and semi-arid regions by synthesizing findings from studies conducted in Iran and elsewhere. The review covers the fundamental principles of soil spectroscopy, sample preparation, spectral preprocessing techniques, multivariate modeling approaches, and recent advances in multi-source data fusion. The reviewed studies demonstrate that Vis–NIR spectroscopy provides satisfactory predictions for several soil properties, including soil organic matter, texture, moisture, electrical conductivity, and calcium carbonate, particularly when coupled with advanced machine learning algorithms. However, predictive performance varies widely in arid and semi-arid soils due to the complex effects of high carbonate, gypsum, and salinity contents, mineralogical heterogeneity, and differences in spectral preprocessing, calibration strategies, and modeling techniques. The review further indicates that variations in model performance are primarily associated with the quality and representativeness of reference datasets, the availability of regional spectral libraries, and site-specific calibration rather than the choice of prediction algorithm alone. Consequently, developing regional spectral libraries, standardizing spectral measurement protocols, and integrating Vis–NIR data with remote sensing and environmental covariates are identified as key priorities for improving model transferability and practical implementation. Overall, Vis–NIR spectroscopy has considerable potential to support precision agriculture, sustainable soil management, and digital soil mapping in arid and semi-arid environments, provided that region-specific calibration and robust validation procedures are adopted.