Soil analysis is a crucial component of agricultural production, as it provides information about the nutrient content in soil to recommend fertilizer guidelines and make informed decisions to optimize crop yields and promote sustainable agriculture. Near-infrared (NIR) reflectance spectroscopy holds great promise as an analytical tool with several advantages, such as rapid, nondestructive, and without the need for chemicals, making it a promising alternative to traditional laboratory methods to determine nutrient contents in soil. This study used two multivariate techniques partial least square regression (PLSR) and locally weighted PLSR (LWPLSR) to construct calibration models to relate NIR spectra to the total nitrogen (total N), available phosphorus (available P), extractable potassium (extractable K) and extractable ammonium (NH4+) contents in soil samples. The results showed that the LWPLSR model outperformed the PLSR model. The best predictions were obtained using LWPLSR model for total N (R2 = 0.87 and RMSEP = 0.133), available P (R2 = 0.68 and RMSEP = 9.013), extractable K (R2 = 0.84 and RMSEP = 15.436) and extractable NH4+ (R2 = 0.73 and RMSEP = 6.789). This work demonstrated the potential of LWPLSR coupled with NIR spectroscopy for efficient soil analysis, providing the geographical neighbors of the test sample present in the calibration database.