Near-infrared (NIR) spectroscopy combined with the partial least squares (PLS) regression was used to establish a rapid no-reagent quantitative analysis method for the content of zinc in the farmland soil samples from Pearl River Delta. The modelling stability was discussed based on varied partitioning of the calibration and prediction sets. The overall scanning region (400－2498 nm) was separated into the visible (400－780 nm), the short-wave NIR (780－1100 nm), the long-wave NIR (1100－2498 nm), and the overall NIR (780－2498 nm) regions. By comparison and validation, the results indicated that the long-wave NIR region had the best prediction accuracy and stability, and the optimal number of PLS factors was 8, the root mean square error (V-SEP) and the correlation coefficient of prediction (V-R_P) in the validation set were 78.847 mg kg_-1 and 0.731, The results show that long-wave NIR spectroscopy can be applied to the rapid no-reagent quantitative analysis of the content of zinc in soil.