Abstract
Molecular biological tools and cheap computing power now permit the assembly of large databases of neuronal images. This has been done for the main classes of retinal cells. The cells can be classified by traditional or more formal means, which agree that mammalian retinas contain ∼ 12 structurally distinct types of ganglion cell. The structures reflect the underlying physical reality of the cells' differing sets of inputs from other retinal neurons, which act to create ∼ 12 parallel encodings of the visual scene. The central targets and central uses of these signals are only partially understood.