In this paper, the method for identification of the origin of dark tea with visible/near-infrared spectroscopy is introduced by taking dark tea as the research object. Dark tea samples were purchased from five different provinces and the spectral data of the corresponding dark tea samples was collected. The dimension was reduced by principal component analysis and the optimal band was selected. The raw spectral data of dark tea samples were preprocessed by convolution-based smoothing, multivariate scatter correction, transformation on standard normal variate, and detrending. Finally, the new spectra and original spectra processed by different pretreatment methods were used to establish a partial least squares regression model on the prediction of the origin of dark tea. The experimental results show that the modeling effect of the transformation on standard normal variate (SNV) for spectral processing is the best; it has the highest coefficient of determination of the test set (RP2), i.e., 0.9627, and the lowest root mean square error (RMSEP), i.e., 0.0255. The study provides a reference for effectively identifying the origin of dark tea, protecting the dark tea brand, and enhancing the image of the national geographical landmark of the brand.