In this paper, we study the effects of wavelength selection and data pretreatments, including no pretreatment, standard normal variate transformation (SNV), vector normalization, smoothing, first and second derivative transformation, on the discrimination of maize seed varieties. The performance of the pretreatment methods is evaluated on the basis of the two data sets: all-range spectral data and the data of the characteristic wavelengths selected by a standard deviation-based feature selection method, respectively. The correct acceptance rate (CAR) and the correct rejection rate (CRR) are used as the criteria for the discrimination models. The results show that the best model uses first derivative and all-range spectral and using the best model both CAR and CRR for five varieties reach 100%, and the average CAR and CRR attains 98.6% and 98%. The wavelength selection can only improve CGR of SNV and vector normalization models.