Based on nearinfrared reflectance spectroscopy technology, a nondestructive testing and grading device for corn seed vigor was developed, including single granulation device, conveying pipeline, nearinfrared spectroscopy acquisition system, control system and grading device, etc. The seed granulation device consisted of an inclined turntable with holes and a fixed tray. The conveying pipe was connected with the fixed tray, and the end of it was a spectrum collecting unit. The seeds were separated by a single granulation device and then dropped into a spectral collection area through a delivery pipe for spectral analysis and viability determination, and then the classified seeds were classified by a classification device. The perforated disc was used for separating seeds, their working efficiency was key for seed detection and grading efficiency. According to analysis, there were three factors that determined the single granulation efficiency of the single granulation device: the tilt angle of the turntable, and the turntable speed and height of the hole. In order to improve the single granulation efficiency of corn seeds and improve the detection rate, the parameters analysis and optimization experiments were carried out on the single granulation device. The results showed that when the tilt angle of the turntable was 31°, the rotation speed was 05r/s, and the height of the hole was 22mm, the seed single granulation efficiency was optimal, and the single channel can reach 7 grains/s. In order to establish a corn seed vigor detection model, based on the device, spectral data of 100 normal viable corn seeds and 100 artificially aged nonviable corn seeds in the wavelength range of 980~1700nm were collected, and the qualitative discriminant model of seed vigor were established by utilizing the PLS-DA in different methods. When the number of principal factors was 5, the modeling comparison results under several different processing modes showed that the modeling effect under SG-smooth preprocessing was optimal. In the calibration concentration, there were one discriminant error of viable seeds and one discriminant error of nonviable seeds, and the discriminant accuracy rate was 987%. All the viable seeds in the predicted concentration were correctly discriminated, and the nonviable seeds had two judgment errors, the total discriminant accuracy was 96%. The stability and accuracy of the device were tested by selecting 100 seeds. The total accuracy of seed vigor prediction was 97%. The results indicated that the selfdesigned corn seed vigor particlefree nondestructive testing grading device had higher single granulation efficiency and stable spectrum acquisition. It was feasible to perform realtime nondestructive detection and grading of corn seeds.