Biomass energy is being industrialized rapidly in China in recent years, whereas, research on energy grass is still in primary stage. Only if near-infrared spectroscopy mode was constructed which was used to predict the lignin, cellulose and hemicellulose contents in energy crop, the varieties screening, performance evaluation and on-line control of industrialization would be facilitated. In this study, the prediction model for quality indices (cellulose, hemicellulose, lignin and ash) of four energy grass (Miscanthus) was built using Fourier transform near-infrared (FT-NIR) spectroscopy combined with partial least squares regression (PLSR) , and the impacts exerted by particle size on the model were also revealed. The results showed that (1) the root mean error of cross validation (RMSECV) of cellulose, hemicelluloses and lignin contents were 1.35% (R2=0.88), 0.39% (R2=0.91) and 0.35 (R2=0.80), respectively in stalk and 0.72% (R2=0.88), 0.85% (R2=0.85) and 0.44 (R2=0.87), respectively in leaf. The model showed good performance in prediction of corresponding contents in unknown samples, however, no satisfying performance in ash content. (2) Both 2 mm and 0.5 mm grades of particle size can meet accuracy requirements of the model. But considering the time and labor cost, 2 mm grade was suggested for model building.