Abstract: For rapid determination of calorific value of clean coal, the near-infrared spectra of 120 clean coal samples are collected. After the abnormal spectrum, if any, is eliminated using studentized residual method, quantitative analysis models of partial least squares regression (PLSR) and principal component regression (PCR) are established, and their modelling effects are compared with different spectral preprocessing methods. As evidenced by result of study, though both models are good in modelling effect, the PLSR model is comparatively superior to the PCR model in performance; and through standard normal variety preprocessing, the PLSR modelling effect can be further optimized with the correlation coefficients of calibration and prediction being up to 0.96 and 0.91 respectively and their mean square errors being as low as 0.001 7 and 0.003 respectively.