Spectroscopy techniques are powerful tools for the rapid identification of traditional Chinese medicine because they provide chemical information with no sample preparation. In this study, a rapid and reliable approach was proposed to differentiate Pinellia ternata from adulterated P. ternata, processed P. ternata, and adulterated processed P. ternata by mid-infrared (MIR) and near-infrared (NIR) spectroscopy coupled with a partial least squares-discriminant analysis (PLS-DA) algorithm. One-hundred sixty-five batches of P. ternata, adulterated P. ternata, processed P. ternata, and adulterated processed P. ternata samples were collected and prepared. All of the samples were characterized by MIR and NIR spectra. The PLS-DA was first applied to build the discriminant model on the individual data matrices. Next, the data matrices coming from MIR and NIR spectra were fused at the low-level and mid-level, and PLS-DA models were built on the fused data. The classification accuracy, sensitivity, and specificity were calculated to evaluate the PLS-DA models. The results showed the use of mid-level fusion strategy, in particular, integrating latent variables from different spectral data matrices, allowed the correct discrimination of all samples in the training and testing sets. In the case of mid-level fusion with latent variables, the accuracy of the PLS-DA model was 100%, and the sensitivity and specificity of the PLS-DA model were all 1. The present discriminant model can be successful to differentiate P. ternata from adulterated P. ternata, processed P. ternata, and adulterated processed P. ternata. This study first provides a new path for the quality control of P. ternata.