Partial least squares method（PLS） was used to establish a method to rapidly predict the air-dry density of Quercus variabilis by near infrared（NIR） spectroscopy. The wood densities of 131 samples from the different radial positions in four trees were determined by referring to GB/T 1933—2009 "Method for determination of wood density", and their NIR spectra were collected. The results showed the effect of model based on NIR spectra taken from the transverse surface was the best. Multiple modeling sample sets were selected by gradient method and multiple PLS models were established. The correlation coefficients of calibration and prediction models were both greater than 0.9. The research showed that it was feasible to predict the wood air-dry density of Quercus variabilis by NIR spectroscopy.