Abstract: A quantitative identification model for testing the purity of Xianyu335 hybrid maize seed was built by near infrared reflectance spectroscopy (NIRS) with quantitative partial least squares (QPLS). By grinding and mixing maize hybrid seeds of different years and sources with their female parent seeds, 123 samples were obtained with a 0.5% gradient and purity within the range of 80%-100% (three replicates of every year and every source in each gradient) and the spectra of the samples were collected. The results showed as following: through implementation of scatter correction pretreatment, the wave number range of 4 000-8 000 cm-1 was appropriate for modeling (calibration sets: validation set = 3:1); the internal cross coefficient of determination (R2) for the calibration set reached 96.06%; the standard error of calibration (SEC) was 1.18%; and the Average absolute relative deviation (AARD) was 1.03%. Further, the R2 for the validation set was 95.02%; the SEC was 1.28%; and the AARD was 1.12%. Results of using different ratios of the modeling samples and testing samples showed that the R2 of the calibration set and validation set were all greater than 94%, indicating the feasibility and the stability of NIRS to quantitatively determine the purity of maize hybrid.