The management of soil nutrients is essential for sustainable agricultural production. The time required for determination of soil nutrients and the high cost per sample are problems attributed to traditional laboratory analyses that limit the adoption of precision agriculture techniques. Such problems arise because the sample density that is required to obtain soil fertility maps is greater than that required by conventional agricultural management. The use of radiometric sensors combined with a diffuse reflectance technique is quick and costs less than the conventional soil analysis performed in laboratory. However, the construction of robust models for the prediction of soil chemical proprieties based on spectral data requires samples with standardized physical characteristics. Thus, the development of a method to determine the soil levels of Phosphorus (P), Calcium (Ca), Magnesium (Mg) and Potassium (K), based on multivariate regression and using spectroscopy in visible and near infrared was the main objective of this work.Ion exchange resins were used as soil nutrient extraction vehicle. Of these, spectra were obtained in duplicates by diffuse reflectance using a FieldSpec® HandHeld 2 ™ spectroradiometer. Standard soil chemical analysis was used as reference method. The models were constructed using the ordered predictors selection (OPS) and partial least squares regression (PLSR) method. The models were evaluated based on the coefficient of determination, the standard deviation ratio with the standard error of prediction (RPD) and the relative error percentages. From the determination of P, Ca, Mg and K by the reference methods and the prediction of the models, spatial distribution maps of these elements were constructed for a coffee crop field. These maps were classified into management zones (ZM’s) and the comparisons between them were made by the coefficient of agreement Kappa. Coefficients of determination greater than 90%, RPD higher than 2.20, and relative error percentages lower than 25% were obtained using the developed models. For the analyzed ions, no significant difference was observed between the developed models and the reference method at the 5% level (p> 0.05). The concordance between the maps (measured and predicted) indicates that the contents of P, Ca, Mg, and K provided by the PLS-OPS models based on diffuse reflectance of ion-exchange resins are reliable. Them, to calculate fertilizer doses can promote sustainability in agriculture, optimizing the production system.