Abstract: Data of 163 groups of coal quality was used in this study and principal component analysis was carried out to acquire the first three principal components of ultimate and proximate analysis of coal. Relationship of coal calorific value and the three was researched by regression analysis. Prediction models of coal calorific value were built and adaptability of the models was tested. The results showed that prediction models of coal calorific value acquired by the principal component - linear regression analysis have good adaptability.