The seed purity is an important indicator of seed quality. The paper proposes a visual identification method of corn seed based on the near-infrared (874~1 734 nm) hyperspectral image technology. Hyperspectral image data of 4 cultivars of a total of 384 corn seed samples will be acquired. Then 288 of samples are to be selected randomly as the calibration set, and the remaining 96 samples will be used for the prediction set. After inspection of the near-infrared spectral curves, 7 effective wavelengths (EWs) are to be selected by successive projection algorithm (SPA). And then 7 EWs of the calibration set will be used as input to build a partial least squares (PLS) model. Good results are to be obtained with RC=0.917 7, RMSECV=0.444 2; RCV=0.911 5, RMSECV=0.459 9. And the total identification rate of the developed PLS model will be 78.5% for the calibration set and 70.8% for the prediction set. Finally, average spectral data of each corn seed in a hyperspectral image will be extracted by image process technology, and used as input of the developed SPA-PLS model. In the produced identification map, different colors are to be used to represent different predicted cultivars. 3 mixture samples of corn seeds will be identified, and help to achieve satisfied visual effects. The result indicates that, by means of the visual identification technology we could intuitively observe the distribution of corn seeds of different cultivars in mixture samples. The research provides help for the identification and screening of seeds in agricultural production.