With the aim to solve the accurate rate short of reflection imaging technology to simultaneously detecting internal and external defects of potatoes, a nondestructive test technology based on transmission imaging and machine vision technology was proposed. It is concluded that the combination of hill climbing method and region growing method is the optimal image segmentation method for transmission and reflection images of potato by studying image preprocessing methods. Partial least squares-support vector machine (PLS-SVM) method was employed to establish the potato defects recognition model for transmission and reflection images of potato. In the potato internal defects detection, the classifying correct rates of the transmission and the reflection imaging technology are 96.30% and 59.26% respectively; in the potato external defects detection, the classifying correct rates are 94.20% and 89.86% respectively; in the simultaneous potato internal and external defects detection, the classifying correct rates are 95.83% and 81.25% respectively. The research results show that the transmission method is better than the reflection method in detecting potato internal and external defects alone, or in detecting the internal and external defects simultaneously.