A model for determining empty grain content in rice seed examination was established based on near-infrared spectroscopy to rapidly and effectively detect empty grains in rice seeds. Firstly, rice samples with different empty grain contents were prepared, and their near-infrared spectral data were collected. To improve the discrimination accuracy of the model, two different combinations of preprocessing methods, including Savitzky-Golay smoothing (SG)+multiplicative scatter correction (MSC)+polynomial baseline correction (PBC) and Savitzky-Golay smoothing (SG)+standard normal variate transformation (SNV)+polynomial baseline correction (PBC) were selected for noise reduction. Besides, three methods of sequential projection algorithm (SPA), competitive adaptive reweighted sampling (CARS), and principal component analysis (PCA) were used to extract the characteristic wavelength variables of the preprocessed spectra, thereby reducing the impact of redundant information in the spectra on the model computation speed and prediction accuracy. Then, based on support vector machine (SVM), K-nearest neighbor algorithm (KNN), decision tree (DT), linear discriminant analysis (LDA), partial least squares discriminant analysis (PLS-DA), and naive Bayes (NB), 6 different identification models for empty grain content of rice seeds were established. Experimental results show that after SG+SNV+PBC preprocessing, the performance of the identification model is better than that of without preprocessing and SG+MSC+PBC. The 158 bands were selected based on the CARS combination SG+SNV+PBC preprocessing band selection. The KNN model established using the selected bands has a better prediction effect, where the testing set identification accuracy of the KNN model could reach 98.47%. The research indicates that near-infrared spectroscopy technology provides a feasible method for discriminating rice seed husk content grades, which provides theoretical support for the non-destructive testing of rice seed quality.