Soil analysis plays a crucial role in precision agriculture, where alternatives or complementary methods to traditional laboratory analysis are needed to reduce costs and processing times. This study evaluated models from different devices for estimating soil organic carbon (SOC) using visible near-infrared (Vis-NIR) spectral data and examined the predictive performance of these models across diverse soil types and land uses. A total of 266 soil samples were collected at various depths from two dehesa farms. Soil reflectance spectra were measured using a LabSpec 5000 spectrophotometer with a contact probe and a Muglight accessory. SOC concentration was determined using the Walkley & Black method. Model prediction accuracy was assessed through metrics including the coefficient of determination (R²), residual predictive deviation (RPD), root mean squared error (RMSE), and range error ratio (RER). Cross-validation demonstrated strong predictive accuracy for SOC, with R² and RPD values exceeding 0.95 and 4.54, respectively, and RER values surpassing 20. Although external validation metrics were more conservative, they still showed excellent RPD indices above 3.12, with no significant difference between devices. Both the Muglight and contact probe yielded low RMSE values (0.222 vs. 0.244) and high R² values (0.90 vs. 0.89). These findings indicate that both devices can reliably predict SOC, with the contact probe offering the added advantage of faster spectrum recording compared to the Muglight.