Abstract: The near-infrared spectroscopy analysis model of cottonseed oil content has been set up, which is non-destructive, fast and efficient, to assay the oil content of cottonseed in large quantities faster and to improve the standard and efficiency of this analysis. A better regression equation was established by using different spectral mathematical pretreatment and spectral reflectance mode, with 218 upland cottonseeds as test sample which is naked cotton, a better regression equation for oil content by near infrared spectroscopy (NIRS) was established through the methods of spectra mathematic pretreatment and different spectra reflection treatment. The results showed the best regression equation was obtained by the mathematics pretreatment“2, 4, 4, 1”for spectroscopic data, and transformation of Standard Normal Variant (SNV). It had the highest regression squared of the calibration (RSQ=0.9773) and verification squared of the calibration (1-VR=0.9646), and the same time, it had the least regression standard error of the calibration (SEC=0.6977), and the least cross-validation standard error of the calibration (SECV=0.8714). There was a correlation coefficient 0.9882 between the data of oil content obtained by the chemical and NIRS method by 30 verified samples, only with a deviation of 0.65%. The regression equation by NIRS could measure the oil content of cottonseed accurately; it could be applied in practice. The average cottonseed oil content of hybrid varieties from the Yangtze River Basin was 28.05%, ranging from 26.50% to 30.05% in majority, it had a large potential for improvement of the cotton varieties.