Perception, 1981, volume 10, pages 421 -422 The frog ganglion cell: not a feature detector and not a monkey cortical cell Donald C Hood Department of Psychology, Columbia University, New York, NY 10027, USA James Gordon The Rockefeller University, New York, NY 10021, USA Received 3 March 1981 There are two fundamental misconceptions about frog ganglion cells that have been perpetuated in most recent textbooks. A large number of texts written for courses on perception, S and P, and physiological psychology use the frog ganglion cell as an example, par excellence, of a feature detector. Many go further and compare the processing done in the frog retina with that carried out in the cat and monkey cortex. The truth is, the properties of the frog's ganglion cells have never fitted the generally accepted definitions of a feature detector. Further, their response properties have recently been shown to be similar to those of other vertebrate ganglion cells. As generally used the term 'feature dectector' refers to a cell that is most sensitive to a specific spatiotemporal aspect (the feature) of the environment. For example, an 'edge detector' is most sensitive to an edge situated at a particular orientation. Often there is a second aspect to the term which refers to what the cell's response 'means' to the animal. In the case of an edge detector, if the cell responds, the animal perceives an edge. Neither aspect of the above definition can be used rigorously or consistently to describe the properties of frog ganglion cells as detailed by Lettvin and his colleagues (Lettvin et al, 1959; Maturana et al, 1960). The feature names applied to the different fiber classes do not necessarily describe the aspect of the stimulus to which these fibers are most sensitive. For example, class 1 cells, or 'edge detectors', do not respond best to edges but to moving spots. These cells were probably called edge detectors because they were the only cells which gave any response to a standing edge. Although this is a reasonable alternative definition of the term 'feature detector', it cannot be applied to other cell classes. Class 2 (the so- called 'bug detectors' or 'convex-edge detectors') respond best to a moving dark spot. In fact, a moving spot is the most effective stimulus for nearly all frog ganglion cells. Consequently it is not surprising that, although class 2 cells are called 'bug perceivers' by Lettvin et al, class 3 cells are called 'fly detectors' by Barlow (1953). Although alternative and reasonable definitions of a feature detector can be applied to some of the Lettvin et al classes, we cannot find one definition that will generate all the class names used (see Gordon and Hood 1976, pp. 56-59 for a more complete discussion). The frog ganglion cells are often described as if they are functionally equivalent to monkey cortical cells. This view can be traced to the misconception of the frog ganglion cells as feature detectors discussed above and to the belief in the 1960s that the mammalian retina, in particular the cat retina, had only two types of ganglion cells, as opposed to the five or six types found in the frog. It is now generally accepted that the cat retina contains perhaps twelve types of ganglion cells, including several types of on-center or off-center X and Y cells, as well as a number of subcategories of W cells (Rodieck 1979). If one studies frog retinal ganglion cells with a set of stimuli designed to measure quantitatively spatial and temporal summation, stimuli which have also been used on the cat and monkey, then it