A linear combination of Gaussian functions at various scales is being suggested as a suitable model for the human visual system. It reduces to the DOG (difference of Gaussian) model at the most primitive level of processing. The model is actually equivalent to the experimentally observed receptive field profiles that can be fitted by various even order derivatives of Gaussians, the order being determined by the number of Gaussians in the linear combination, once again reducing to the DOG-LOG (Laplacian of Gaussian) equivalence at the most primary level of visual signal processing. The role of amacrine cells in the retina is explained in this light and the inherent multi-scale property of the model is looked upon as a suitable mechanism for enabling a unified representation for the various classes of retinal ganglion cells differing in their receptive field profiles.