Near infrared spectroscopy was applied to rapidly predict density, modulus of rupture and tensile strength parallel to grain of neosinocalamus affinins. Backward interval partial least squares (BiPLS) was used to find the most informative spectrum ranges, and build models based on raw spectra and pretreated spectra, including first derivative spectra, second derivative spectra, Savitzky-Golay smoothing spectra and standard normalized variate spectra. And partial least squares (PLS) models were also developed in the whole wavelength range 350-2500 nm. The results show that compared with PLS models, BiPLS could effectively find the optimal spectrum regions and improve the predictive ability of models. The optimal models of density, modulus of rupture and tensile strength parallel to grain were obtained through BiPLS method that separated the whole spectra pretreated by standard normalized variate, second derivative and first derivative respectively into 20, 30 and 20 intervals. And the prediction models of density, modulus of rupture and tensile strength parallel to grain had correlation coefficient (r) 0.85, 0.88 and 0.88, as well as root mean standard error of prediction (RMSEP) 0.0524, 0.0185 and 0.0292, respectively. The relation between NIR predicted values and actual values was good in all cases. Therefore, the experimental results demonstrated that NIR spectroscopy was promising for predicting the physical and mechanical properties of neosinocalamus affinins.