Hyperspectral imaging is concerned with the measurement and the analysis of images acquired in contiguous spectral bands over a given spectral range. The combined use of available spectral and spatial information for object detection seems essential in many fields of application (characterization of urban areas, agriculture, etc.). The objective of this study was to apply a spectral-spatial approach named butterfly (Gorretta, 2009), for the problem of vitreous durum wheat kernels classification. The spectral-spatial results were compared with the one of a purely spectral classification method, i.e. the Partial Least Squares Linear Discriminant Analysis (PLS-LDA).