The information on coal species provides technical support for evaluation of coal quality and import and export tax. Traditional coal identification methods require the determination of indicators such as dry ash-free volatile matter, the light transmittance of low-rank coal, bonding index, the gross calorific value on the moist ash-free basis of coal samples and other indicators, with large energy consumption and long detection cycle, which is not conducive to the rapid customs clearance at ports. Due to the advantages of no chemical reagents consumption, and fast and low cost, the research on coal identification by near-infrared spectra has attracted extensive attention. However, there has not been any application for the identification of coal from different sources in the world so far, and the correlation between NIR spectral characteristics of cal and coal species remains to be explored. This research collected 410 representative samples of imported coal from 9 countries, including Australia, Russia and Indonesia, etc. involving lignite, bituminous coal and anthracite. By analyzing the near-infrared spectrum, it is found that the differences in NIR spectra of different coal species mainly focus on absorbance, spectral slope and characteristic peak. Combining sample composition information, X-ray diffraction analysis and near-infrared spectra to analyze the reasons for these differences shows that the NIR absorbance is positively correlated with the fixed carbon content in coal, and the spectral slope is negatively correlated with this the aromatization of coal. The increase of coal aromatization leads to the increase of the absorption coefficient in the long-wavelength direction and the decrease of the spectral slope. Spectral characteristic absorption peaks are mainly the characteristic information of water and hydrogen-containing groups of organic substances, and the intensity of characteristic peaks depends on the content of water and volatile matter in coal. Principal component analysis (PCA) was used for data dimension-reduction, and the spectral variables were reduced from 1 557 to 394. The first 10 principal components were discriminated step by step, and PC1, PC2, PC3, PC4, PC6, PC7, PC8, PC9 and PC10 were selected as input variables to establish the Fisher discriminant analysis model for coal species identification. The verification accuracy of modeling sample was 98%, the cross-validation accuracy was 97.8%, and the verification accuracy of the test sample was 99.1%. PCA load diagram shows that PC1 and PC2 are mainly related to the volatile content of coal, followed by moisture content. The correlation between the discriminant function 1 (57.7%) and PC1 was the strongest, and the correlation between the discriminant function 2 (42.3%) and PC2 was the strongest, which indicated that the difference between volatile content and moisture content in different coal species was the internal basis for the identification of coal species by NIR.