Abstract: 【Objective】The traditional iodine colorimetry method requires complex and time-consuming pretreatment and experimental preparation. The near-infrared detection method not only has the advantages of simple operation and rapid detection, but also does not consume reagents, samples and standards, which greatly reduces the detection cost. This paper aimed to establish a near-infrared detection model suitable for the starch in Jiangxi flue-cured tobacco based on Jiangxi flue-cured tobacco samples. 【Method】The near-infrared spectra and corresponding continuous flowing analyzer detection data(iodine colorimetry) of 650 flue-cured tobacco samples from 21 districts and counties in Jiangxi Province were collected, then the detection model was established by using different regression methods and data processing methods with the spectral data and chemical analysis data in one-to-one correspondence, and the best model parameters were determined by comparing their root mean square(RMS) errors and correlation coefficients. 【Result】A starch content prediction model suitable for Jiangxi flue-cured tobacco was established. The RMS errors of calibration and the prediction were 0.407 and 0.490 respectively, and the correlation coefficient was 96.52%.The model was verified by external samples and the errors of 95% of the samples were within 10%. 【Conclusion】The model reduced the detection cost and improved the analysis efficiency, which could be used for the rapid detection of the starch content in Jiangxi fluecured tobacco samples and replace the traditional iodine colorimety method to a certain extent .