Near infrared spectral analysis technique can be used to detect samples quickly and nondestructively, which is playing an increasingly important role in people’s production and life. The support vector machine is a commonly used method for building qualitative analysis models. It separates two kinds of samples by finding the optimal classification hyperplane. In the case of small samples, the support vector machine method has its unique advantages. The principal component analysis is a commonly used method to reduce the dimension of data. After the dimension is reduced by this method, the data is used as input variables of the support vector machine method. The model can be simplified and the accuracy of discriminating by the model can be improved in this way. So the support vector machine based on the principal component analysis (PCA-SVM for short) is suitable for establishing the qualitative analysis model of near infrared spectroscopy. The multi-model method is a modeling method seldom used by people. The model established by this method usually has good stability. The multi-model method is successfully combined with the PCA-SVM method to form a new method in this paper. With cotton and nylon blended, cotton and polyester blended textiles being taken as an example, a qualitative analysis model of near infrared spectra of these two types of textile samples is established by the new method. In modeling, the spectral data are divided into 4 groups according to the wavelengths. A sub model is established with each group of spectral data. The final prediction results are obtained by weighted average of the output values of the sub models. The information contained in the spectral data can be used more fully in this way. In order to facilitate the comparison of different methods, the aforementioned calibration set and validation set are used. A qualitative analysis model of near infrared spectra of these two types of textile samples is also established by using the PCA-SVM method in the paper. The cross validation of the prediction results show that the mean value of the correct rate of discrimination by the model built with the new method is 85.49%, the standard deviation of the correct rate of it is 0.066 7, and the mean value of the correct rate of discrimination by the model built with the PCA-SVM method is 83.34%, the standard deviation of the correct rate of it is 0.109 6. Since the mean value 85.49% is higher than the mean value 83.34%, the classification effect of the model built by the new method is better than that built by the PCA-SVM method. Since the standard deviation 0.066 7 is much smaller than the standard deviation 0.109 6, the stability of the model built by the new method is obviously higher than that built by the PCA-SVM method. The prediction effect of the model built by the PCA-SVM method is greatly influenced by the composition of the calibration set. But the prediction effect of the model built by the new method is relatively stable. Sorting and recycling waste textiles can save a lot of textile raw materials. However, manual sorting is inefficient and costly. Classification of textiles by using the method of near infrared spectra analysis is proposed in this paper, which lays a certain foundation for large-scale fine sorting and grading of waste textiles. The new method put forward in the paper is also expected to be used for classification of some other types of samples.