Automatic visual grading of seed lots with a high density of touching grain kernels is a challenging problem. The structure tensor is a simple and robust descriptor that was developed to analyze textures orientation. Contrarily to segmentation methods which rely on an object based modelling of images, the structure tensor views the sample at a macroscopic scale, like a continuum. Thanks to this tool, it is possible to extract useful information on the orientation of grain kernels even in a bulk. Knowing the rough orientation of a grain kernel could provide an initialization for segmentation techniques. The comparative results with ground truth orientations on four different grain kernels demonstrate the ability of the structure tensor to detect the seed orientations.