Abstract: Savitzky-Golay smoothing and wavelet transform were applied to the near infrared spectrum(NIRS)of oak(Quercus mongolica Fisch. ex Ledeb)wood for the noise elimination. Denoising effects were discussed by switching the pretreatment order of the two methods. Partial least squares regression(PLS)and principal component regression(PCR)were used for model development of oak wood air-dry density prediction. The results showed that the best denoising effects were associated with spectrum pretreatment of SG smoothing followed by wavelet transform with three smoothing points and two db5 wavelet decomposition layers. The signal to noise ratio(SNR)was 18.546 and the root mean square error of calibration was 0.04. For the calibration model of NIR-based oak wood air-dry density prediction developed with PLS, the determination coefficient(R²)was 0.902 for the calibration model with correct root mean square error reduced by 35.32% compared to the model developed with raw spectrum. R² for prediction was 0.860 with residual prediction deviation of 2.67. This study indicated that NIR could be used for rapid determination of oak wood air-dry density, and determination of reasonable processing parameters as well as modeling methods could effectively improve the model accuracy.