In order to detect the external quality of potato quickly, the hyperspectral imaging technology was used. Potato with germination and other three kinds of common defects were studied. The partial least-squares discriminant model were built after different pretreatment methods for spectral data processing. The results showed that pretreatment method of SNV was the best. 13 and 9 feature bands were selected after using successive projections algorithm (SPA) and weighted weight method (WWM) for spectral data preprocessed. The support vector machine (SVM) discriminant model were established for both SPA and WWM. Our results also showed that the two methods to predict the set of discriminant accuracy reached 100%. WWM-SVM discriminant model of calibration set of cross validation rate was 99.5%, higher than that of the SPA-SVM discriminant model. The study demonstrated the feasibility of using hyperspectral imaging technology combined with WWM-SVM and SPA-SVM for potato external quality grading.