Botanical origins of propolis are significant factors affecting biological and pharmacological activities because of different components in propolis. Until now, the determination of propolis botanical origins is mainly based on different varieties and the content of the compositions with great limitations. Therefore, it is important to discriminate different botanical origins of propolis quickly and accurately. In this study, Near-infrared (NIR) spectra of propolis varieties based on principal component analysis mahalanobis distance (PCA-mahalanobis distance) model and canonical discriminant analysis model were built for the classification of three botanical origins (poplar propolis, brich propolis and rubber propolis). The models were built based on the optimal pretreatment method and bands of first derivative + Savitzky-Golay (7) filter and 4500~12 000-1, which were selected in advance. After the principal component analysis, the correct classification rates of calibration sets and validation sets in analysis mahalanobis distance models were 93.62% and 82.61%, respectively. The discrimination rate and the cross-validation rate of canonical discrimination models were 91.4% and 88.6%, respectively. Therefore, NIR spectroscopy with chemometric methods is not only feasible but also practical for rapid and accurate identification of varieties of propolis.