A near-infrared spectroscopy (NIRS) model of single brown-rice kernels moisture content is established and optimized by applying different preprocessing treatment and variables selection methods. Results show that in the following three ranges, from 5292 cm－1 to 5616 cm－1, 7236 cm－1 to 7600 cm－1, 7884 cm－1 to 8208 cm－1, the predictive ability of single brown-rice kernels moisture content partial least squares (PLS) model established by standard normal variation spectral pretreatment is optimal. Its determination coefficient is 0.98 and prediction root mean square error is 1.01%. The multivariate linear regression model(MLR) of single brown-rice kernels moisture content at 5492.56, 7158.84, 8285.12 cm－1 has the least variables and better prediction ability, whose determination coefficient is 0.9661 and prediction root mean square error is 1.1137%. Results show that near infrared spectroscopy can be employed to determine single brown-rice kernels moisture content rapidly and accurately.