Rice is staple food for people in the large part of the world, especially in Asian countries. The major chemical constituent contents of rice are moisture, protein and starch (amylose and amylopectin). Those chemical constituents associate with eating quality of rice. Near-infrared (NIR) spectroscopy is one of the non-destructive methods for determining grain chemical contents. Moisture and protein contents can be measured with high accuracy using NIR spectrometer at rice grain elevator in Japan. However, the accuracy to measure amylose content is not sufficient. Thus, there is a strong need for the highly accurate measurement of rice amylose content using non-destructive method. The overall objective of this study was to develop non-destructive techniques to determine rice amylose content for practical use at rice grain elevator. Amylose content measurement was performed using an auto-analyzer for reference (chemical) analysis. Spectra data of rice were obtained using an NIR spectrometer with a wavelength range of 850 to 1048 nm. Calibration model to determine amylose content was developed using non-waxy Japonica-type rice samples (calibration set, n=974, including 14 cultivars) grown in Japan from 2008 to 2016. Partial least squares (PLS) regression analysis was used to develop calibration model. The accuracy of the model was validated using other rice samples (validation set, n=95, including 10 cultivars) grown in 2017. There are two cultivar groups of non-waxy Japonica-type rice produced in Japan. One is ordinary amylose content rice cultivar and the other is low amylose content rice cultivar. The accuracy of determination of rice amylose content was improved by increasing production years of calibration sample set. When one calibration model to determine amylose content was developed using all cultivars together, validation statistics were shown: coefficient of determination (r²) was 0.72, bias was -0.04%, standard error of prediction (SEP) was 0.92%, and ratio of SEP to standard deviation of reference data (RPD) was 1.90. To improve the accuracy of determination of rice amylose content, two calibration models were developed using ordinary amylose content rice cultivars or low amylose content rice cultivars, respectively. As a result, the accuracy of two calibration models was better than that of one calibration model. Validation statistics of the two calibration models were shown: coefficient of determination (r²) was 0.93, bias was 0.01%, standard error of prediction (SEP) was 0.64%, and ratio SEP to standard deviation of reference data (RPD) was 2.76. The production year of validation set (2017) was different from those of calibration set (2008 to 2016). This is the same condition as practical use of this non-destructive method at rice grain elevator. The result obtained in this study indicated that the two calibration models enables non-destructive determination of rice amylose content at rice grain elevator.