Last-generation GAN models allow to generate synthetic images which are\nvisually indistinguishable from natural ones, raising the need to develop tools\nto distinguish fake and natural images thus contributing to preserve the\ntrustworthiness of digital images. While modern GAN models can generate very\nhigh-quality images with no visible spatial artifacts, reconstruction of\nconsistent relationships among colour channels is expectedly more difficult. In\nthis paper, we propose a method for distinguishing GAN-generated from natural\nimages by exploiting inconsistencies among spectral bands, with specific focus\non the generation of synthetic face images. Specifically, we use cross-band\nco-occurrence matrices, in addition to spatial co-occurrence matrices, as input\nto a CNN model, which is trained to distinguish between real and synthetic\nfaces. The results of our experiments confirm the goodness of our approach\nwhich outperforms a similar detection technique based on intra-band spatial\nco-occurrences only. The performance gain is particularly significant with\nregard to robustness against post-processing, like geometric transformations,\nfiltering and contrast manipulations.\n