The traditional chemical methods to measure the volatile oil content of zanthoxylum bungeagum encounter some problems such as long time and low efficiency, so it is difficult to achieve rapid detection. One hundred forty-one samples including 74 zanthoxylum bungeagum maxim and 67 zanthoxylum schinifolium Sieb. et zucc were collected, from many provinces in China such as Shan Xi, Si Chuan, Gan Su, Chong Qing, Yun Nan, etc. Each sample was crushed and sorted to 8 kinds of powder samples according to the particle size of 120-mesh, 100-mesh, 80-mesh, 60-mesh, 40-mesh, 20-mesh, 10-mesh, respectively, including the material retained by the 10-mesh sieve. Then, each powder sample was labeled by one of the following serial numbers: 120, 100, 080, 060, 040, 020, 010 and 000. For each sample, the NIR spectra of 8 different kinds of particle size powders were measured using a Bruker MATRIX-I FT-NIR spectrometer. Then, the 8 different kinds of particle size powders of each sample were mixed uniformly. The volatile oil content was measured in each sample according to the distillation stipulated by the Forestry Standard of PRC-Quality Classify of Prickly Ash (LY/T 1652-2005). Based on near infrared spectroscopy technique and partial least squares (PLS), 8 calibration models of predicting volatile oil content were established by 141 powder samples with 8 different kinds of particle size. Experiments indicatd that the model was the best with the powder's particle size of 40-mesh and the determination coefficient (r2(141)) and the root mean square error of cross validation (RMSECV141) were 0.9364 and 0.421, respectively. The model was established by the calibration set with 105 samples with particle size of 40-mesh. Applying the model to the test set with 36 samples, the determination coefficient (r2(36)), the root mean square error of prediction (RMSEP36), the relative standard deviation (RSD36), and the ratio of prediction to deviation (RPD36) were 0.9233, 0.452, 11.66%, and 3.624, respectively. The model, based on the same sample set but optimized by OPUS 5.0, was developed by spectral data pretreatment of the Mean Centering+Vector Normalization in the spectral region of 6 100.1-5 774.2 cm(-1) and 4 601.6-4 424.2 cm(-1). Using the model to predict the test set, r2(36), RMSEP36, RSD36, and RPD36 were 0.9862, 0.192, 4.95%, and 8.517, respectively. The results showed that the model built by samples passed through 40-mesh screen was the best and rapid detection of volatile oil content in zanthoxylum bungeagum by NIR was feasible and efficient.