Abstract: Abstract: The problem of heat-damaged seeds frequently occurs because of improper storage of moist seeds or artificial drying of damp seeds at high temperature, thus influencing its sale and usability. It is therefore vital for seed companies and farmers to identify damaged seed. The current visual method for discriminating heat-damaged seeds is subjective and based on discoloration. However, heat damage does not always cause a color change in kernels while it could cause protein denaturation which may result in NIR absorption differences between native protein and denatured protein. In this study, we investigated the possibility of using near infrared spectroscopy to classify good and heat-damaged seeds. A group of 60 tomato seeds was heat-damaged with high temperature treatment while another group of 60 samples was good seeds without heating treatment. The laboratory self-constructed near infrared spectroscopy system, which measured reflectance spectra from 980 to 1700 nm, was used to obtain single seed spectra . In order to verify the difference of viability between heat-damaged seeds and good seeds, a standard germination experiment for 14 days was conducted after the spectra of samples were done. The germination rate, germination potential and germination index of heat-damaged seeds were significantly lower (p < 0.05?) than that of good seeds, and the average germination days was higher (p < 0.05?) than that of good seeds, which indicated that the viability of heat-damaged seeds was lower than that of good seeds. Then, the samples were divided into calibration set and validation set according to the ratio of 3:1 by Kennard-Stone method which is widely used in the qualitative analysis of spectral data. Methods of partial least squares discriminant analysis (PLS-DA) and support vector machine (SVM) were applied to establish the discriminating models for heat-damaged tomato seeds. The results showed that those two discriminative models could be used to differentiate heat-damaged seeds and good seeds. The total accuracy of each validation set was higher than 96%. For PLS-DA model, total classification accuracy for both the calibration sample set and validation sample set were 100% and 96.7%, respectively when five PLS factors was selected by leave-one-out cross validation. The classification accuracy of the good seeds of the validation set was 100%, but one heat-damaged sample was misjudged as the good sample. SVM model yielded higher classification accuracy than PLS-DA model, which was more suitable for classifying heat-damaged tomato seeds according to near infrared reflectance spectra. The SVM model based on principal component analysis (PCA) of the preprocessed spectral data gave the best result, its classification accuracy of the calibration set and the validation set were 100%. Moreover, the prediction bias of PCA-SVM model was less than that of the PLS-DA model, and the average deviation of the validation set was 0.024, which was more conducive to the stability of the model. The overall results suggest that near infrared spectroscopy technique combined with proposed pattern recognition algorithm is accurate for classification of heat-damaged and sound seeds, and provides a new method for future research about nondestructive testing of seed quality.