Near infrared spectroscopy（NIR）technique was applied to compare the influence factors of Eucalyptus pellita’s air-dry density. Air-dry density of eucalypt wood was tested by direct measurement. After collecting the near infrared reflectance spectra of samples in different section and with different thickness, moisture content and roughness, the NIR spectra were preprocessed with the second-derivative and the regression models were built in certain spectra. The calibration models were established using 50～140 samples with the partial least squares method and validated with external validation method. The results showed that the predicted results were influenced by sample’s section, thickness, roughness and moisture content. The best near infrared spectroscopy prediction model was built under the condition of transverse section, 2～5 mm thickness, 12% moisture content and meticulous roughness of wood.