Microstructure and mechanical properties of heat-resistant steel will deteriorate during the service process. The real-time monitoring of the aging state is of great significance for safe operation and production. In this study, a portable laser-induced breakdown spectroscopy (LIBS) device is used to quickly diagnose the aging grade of T91 steel, while the obtained spectral features are dimensionally reduced and the modeling method is optimized. Principal component analysis (PCA) and linear discriminant analysis (LDA) are used to optimize and simplify the spectral features. Finally, after dimensionality reduction, the data are used to evaluate the aging grade model based on the K-nearest neighbor and the support vector machine (SVM) algorithms. Further, the influence of key parameter selection on the model performance is studied. The results show that the spectral data reduced by LDA can achieve a better clustering distribution and improve the accuracy of the evaluation model. In addition, the LDA-SVM model can achieve 94.58% accuracy, which is the highest among all the mentioned aging grade evaluation models. The result demonstrates that the modeling method can efficiently realize the aging grade evaluation of T91 steel based on portable LIBS.