Abstract: Moisture of chestnuts is one of the key indicators affecting their storage and processing. The aim of this study was to evaluate the feasibility of near infrared reflectance (NIR) spectroscopy to determine the moisture of chestnuts with rapid non-destructive testing. The Sample Set Partitioning based on joint X–Y distances, as a method to improve the model performance, was explored when the calibration and validation subset were partitioned. Partial least squares regression models based on a population of 240 samples were established for chestnuts with and without peel respectively and the effect of different preprocessing, namely First Derivative, Second Derivative, Standard Normal Variety Multiplicative Scatter Correction and Non-preprocessing to the performance of models were compared. The results showed that for both peeled and intact chestnuts, the models developed from spectra after First Derivative preprocessing achieved the optimal performance. The model for peeled chestnuts gave more accurate prediction with 1.58% as the root mean square error of calibration (RMSEC), 1.48% as that of prediction (RMSEP), and 0.92 and 0.90 as the correlation coefficient (R) of calibration and validation subset, respectively. These parameters of the model for intact chestnuts were 2.35%, 2.29%, 0.82 and 0.79, respectively. The overall results indicated that NIR spectroscopy could be applied as a nondestructive and accurate alternative method for the determination of chestnuts moisture during orchards and post-harvest process rapidly.