If the waste textiles are classified, recycled and disposed of according to their components, many textile raw materials can be saved. At present, the manual sorting method is often used in the recycling process of waste textiles. This method is costly and inefficient. Near-infrared spectroscopy analysis is one of the most rapidly developing technologies in the 21st century. It can quickly determine the components of the sample and the content of each component without destroying the sample. Using this technology to analyze the waste textiles and prejudge the components and contents of various components of waste textiles can be helpful for the large-scale fine classification and recycling of waste textiles. In the multi-model method, the final predicted value is obtained by a weighted average of the predicted values of each sub-model. The near-infrared spectroscopy analysis model established by this method generally has good stability. In this paper, taking the nylon content of waste textile samples as an example, a near-infrared spectral analysis model for predicting the nylon content is first established using the multi-model method. The method is as follows: The reflectance vectors are divided into 15 groups according to their wavelengths. A sub-model of near-infrared spectral analysis is established with each data group. The final predicted value of the nylon content is obtained by a weighted average of the predicted values of sub-models. Then, based on the multi-model method, according to the approximately linear relationship between the predicted values and the experimental values of the nylon content, by replacing constants with variables and by standardizing the variables, a new model for predicting the nylon content by near-infrared spectral analysis is presented, and the model is convenient for optimization. After optimization, the parameters of each sub-model are reduced by 6. This can prevent overfitting of the model.The above two models are compared with the common model established by the partial least squares method. The results of cross-validation show that: the average of the goodness of fit of the (optimized) new model is 0.820 7. The average goodness of fit of the model built using the multi-model method alone is 0.769 1. The average goodness of fit of the model built by the partial least squares method is 0.746 7. Therefore, the prediction effect of the model built by the multi-model method is better than that of the model built by the partial least squares method. The prediction effect of the new model is better than that of the other two models. The main innovation of this paper is the establishment and optimization of the new model. The modeling method in this paper is expected to predict the content of other components in waste textile samples.