Cold medications are widely used in daily life, and the quality and safety of the medications are directly related to the personal health of the user. However, with more and more lawbreakers imitating medications for profit and medication regulation looming, it is important to distinguish among cold medicines with similar ingredients. In this paper, the combination of laser-induced breakdown spectroscopy (LIBS) with t-distributed stochastic neighbour embedding (t-SNE) and back propagation neural network (BPNN) was proposed for classification and traceability of cold medications with the same main ingredient, compounded paracetamol and amantadine hydrochloride. Firstly, the elements contained in six extremely similar cold medications were analysed by LIBS. Then, t-SNE, a method of high-dimensional data visualization and dimensionality reduction, presented better classification results than other methods, i.e., principal component analysis (PCA), local semantic analysis (LSA), independent component analysis (ICA), linear analysis (Linear), Isomap and local linear embedding (LLE). However, it has limitations because of the number of datasets. Finally, BPNN was used to accurately trace these cold medications with its highly self-learning and self-adaptive ability, and the recognition accuracy was up to 96.7%. The results show that LIBS combined with t-SNE and BPNN provides assistance in drug classification as well as traceability.