Using 147 southern japonica rice varieties or combinations as test materials, the near-infrared analysis model of rice amylose content was constructed by partial least squares (PLS) method. The results showed that the full spectral band (950-1 650 nm) was the best performance for modeling with the correlation coefficient (R), 0.947 7; prediction standard deviation (SEP), 1.162 3, and calibration standard deviation (SEC), 0.700 2, respectively. With different preprocessing methods, the models are adopted for the full spectrum. The multi-spectral correction (MSC) method showed the best performance. The resulted optimized model had correlation coefficient (R), prediction standard deviation (SEP) and calibration standard deviation (SEC) 0.981 9, 0.100 9 and 0.683 1, respectively. The relative analysis error (PRD) was 3.6. The results of paired T test showed that P=0.356>0.05 (confidence interval 95%) indicating that there was no significant difference between the predicted values of near infrared spectroscopy and the results of chemical values. It is feasible to detect the amylose content of rice by near-infrared spectroscopy.