There is no doubt about how useful and valuable the information provided by the hyperspectral sensors can be. Image segmentation procedures can take advantage of this information to increase the ability for separating different textures in an image. A multiscale approach for segmenting hyperspectral images is presented in this work. The method is based on the recently proposed texem model which has been extended in this work to spaces of high dimensionality. Furthermore, the hyperspectral extension of the texem-based segmentation would be computationally impracticable without a prior step for reducing the dimensionality. Thus, a band selection process based on the mutual information among bands has also been applied. The complete process is particularly useful in applications for remote sensing or quality inspection tasks.