Classification of remotely sensed images is often based on assigning classes on a pixel by pixel basis. Such a classification ignores often useful reflectance information in neighbouring pixels. Open types of natural land cover such as maquis and garrigue ecosystems as found in the Mediterranean region may be classified successfully by methods accounting for reflectance patterns in neighbouring pixels. Classification methods capturing neighbouring pixel information are referred to as contextual classifiers. In this paper a new method, the spatial and spectral classifier or SSC is proposed that combines the advantages of two classification methods based on spectral information and on contextual information from neighbouring pixels. The SSC method starts by dividing a hyperspectral image into homogeneous and heterogeneous regions based on spectral variation of pixels within a kernel. Next, the homogeneous image parts are classified using a conventional per-pixel method. The heterogeneous image sections are classified using a combination of spectral and contextual information. The method was tested and the accuracy assessed using airborne DAIS7915 hyperspectral images acquired over an area in southern France covered by semi-natural vegetation, agricultural fields and open mining activities. Classification accuracy is compared with results of purely spectral classifiers. Results were promising and indicate that the accuracy of the SSC classifier was higher than that of the conventional per-pixel classifiers.