We propose an effective approach for determining the manufacturer of oxytetracycline based on terahertz time-domain spectroscopy and chemometrics. The applied method allowed detection and distinction of oxytetracycline tablets produced by four Chinese manufacturers. Absorption spectra within the frequency range of 0.2–1.6 THz were acquired for analysis, and principal component analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE) were implemented for dimensionality reduction. The obtained data were input into different classifiers, namely, grid-search support vector machines (SVM), genetic algorithm-SVM, particle swarm optimizations (PSO-SVM), decision trees, and random forests. The results indicated that the absorption spectra of oxytetracycline produced by the four manufacturers could be differentiated, and that t-SNE outperformed PCA in terms of dimensionality reduction. The optimal classifier was obtained by combining t-SNE and PSO-SVM, and this approach demonstrated the highest cross validation accuracy (98.33%), an average training set accuracy of 97.72%, and an average test set accuracy of 94.25%. The novel approach described herein achieved rapid classification of oxytetracycline and allowed accurate tracing of the manufacturers.