In forensic scientific evidence identification, the classification and recognition of disposable masks is critical. Therefore, in this study, a classification and recognition method for disposable masks is proposed using Raman spectroscopy and machine learning. First, the Raman spectra of 37 types of disposable masks are gathered from different cities and manufacturers. Then, the data are processed using Savitzky Golay smoothing and normalization algorithms, and the mask categories are classified using principal component analysis method and the Raman spectrum characteristic peak comparison method. Finally, a disposable mask classification and recognition model is constructed based on support vector machine (SVM), Bayes discriminant analysis, and back propagation (BP) neural network algorithm. The experimental results show that the accuracy rates of the training and tests set of the SVM model are 93.3% and 100.0%, respectively, whereas the those of the training and test sets of the Bayesian discriminant analysis model are both 100.0%. Furthermore, the accuracy rates of the training set, validation set, and test set of the BP neural network model are 93.9%, 60.0%, and 60.0%, respectively. Therefore, as the best model for mask classification and recognition, the Bayes discriminant analysis model provides a certain reference for forensic science and technology.