Rapid and nondestructive analysis of sedative drugs plays an important role in forensic science. To demonstrate the potential of classifying benzodiazepine sedative drugs, we examined 81 samples from eight types of benzodiazepine sedatives. Spectral data of each sample were analyzed by attenuated total reflection-surface enhanced infrared spectroscopy (ATR-SEIRAS). Fisher discriminant analysis (FDA) and multilayer perceptron neural network (MLPNN) models were constructed based on the original spectral dataset, first derivative spectral dataset, and spectral fusion dataset. The results showed that there were some differences in the physical and chemical details of different samples. ATR-SEIRAS spectra could reflect these differences, which laid a foundation for the effective classification of different benzodiazepine sedatives. The classification accuracy of the FDA model based on the spectral fusion dataset was the highest (100%), and the classification accuracy of the first derivative spectral dataset and original spectral dataset was 96.3% and 92.6%, respectively. Based on the above datasets, the classification accuracy of the MLPNN model was 97.5%, 96.3%, and 88.9%, respectively. Overall, the results demonstrate that ATR-SEIRAS combined with FDA and MLPNN classifiers can achieve rapid and nondestructive classification of eight types of benzodiazepine sedatives.