In order to achieve the accurate classification of sedative-hypnotic drugs, the infrared spectra data of 92 samples of 3 kinds of sedative-hypnotic drugs including benzodiazepine, phenothiazine, and barbital were collected. Then the classification model was constructed after data preprocessing. The experimental results show that the accuracy of series fusion is better than mean fusion and difference fusion. The random forest model was used for modeling and analysis after the fusion of the fingerprint region characteristic variables of the original spectrum and the first-order derivative spectrum. It could achieve thorough identification of three kinds of sedative-hypnotic drugs, with the accuracy up to 100%. The method is rapid and accurate, which can provide a new method and idea for the examination of hypnotics in judicial identification to a certain extent.