A prediction model was built to estimate the antioxidant property of methanol extracts of 38 crude drugs using Near Infrared Spectroscopy (NIS) with partial least squares (PLS) regression method. In order to enhance the chemical information and reduce data systemic noise, the effect of spectral pretreatment methods on the model were compared according to the correlation coefficients of cross validation (R2) and the root mean square errors of cross validation (RMSECV). Prediction effects of the samples were investigated with the root mean square errors of prediction (RMSEP), and the residual predictive deviation (RPD). The DPPH method was employed for verification. The present results showed that the calibration model was developed by first derivative+vector normalization with the selected spectral region, R2, RMSECV, RMSEP, and RPD were 0.896 0, 4.35%, 3.62%, and 2.38, respectively. The NIS model established in the present study was relatively stable, accurate and reliable for overall evaluation of antioxidant activity of crude drugs. However, it was necessary to improve the precision and the robustness of the model for practice．