Many studies have demonstrated that the primary visual cortex contains multiple functional maps of visual properties (e.g., ocular dominance, orientation preference, and spatial-frequency preference), but as yet no consistent picture has emerged as to how these maps are related to one another. Three divergent, prior opticalimaging studies of spatial frequency are reanalyzed and critiqued in this article. Evidence is presented that a nonstimulus-specific response biased the interpretation of results in previous studies. In addition to reexamining four prior cat experiments, we carried out one new experiment. Through the use of different methods and a careful removal of the nonspecific response, we are led in all instances to a unique view of cortical organization for spatialfrequency preference. In particular, we find little apparent evidence for a columnar organization for spatial frequency. The response recorded by each image pixel may be viewed as arising from an admixture of low-and high-spatial-frequency populations. For most pixels, the ratio of these populations is 1:1.optical imaging ͉ nonspecific response ͉ generalized indicator ͉ function analysis T he retinotopic map relates an element of visual space to an element of cortical tissue. Neurons in the primary visual cortex are known to be differentially sensitive to orientation, , and spatial frequency (SF), k, of corresponding image elements that appear in visual space. ‡ Thus, it may be said that the primary visual cortex maps visual space through a local Fourier analysis (1, 2). The process by which this mapping takes place, although not fully understood, is known to involve information processing by the retina, lateral geniculate nucleus (LGN), and primary visual cortex area V1 (V1) and feedback from higher areas.Casting aside the nature of neuronal circuitry, the functional layout of the primary visual cortex is well established for orientation preference and ocular dominance. Anatomical (3-5) and optical-imaging (6-9) studies have elaborated and extended the columnar view of V1 that emerged from electrophysiology (10, 11). Less can be said of the representation of SF. Extensive electrophysiological recordings (12-17) and optical-imaging experiments have not provided a consistent picture of SF layout. The situation can be further complicated if variations in the temporal properties of the stimuli are included (18).One series of optical-imaging studies (7,(19)(20)(21)(22) advanced the viewpoint that the orientation layout of V1 is organized into high-and low-SF preference columns. From this view, it follows that only a single map is needed to characterize SF. This map contains regions corresponding to high and low SF with opposite signs. Thus high-and low-SF stimuli result in responses of opposite signatures. An obvious shortcoming is that intermediate SF would lead to null or near-null results. On the other hand, Everson et al. (23) reported that SF preference is continuously distributed and that at least two maps are required to depict SF represen...

Relay cells in the mammalian lateral geniculate nucleus (LGN) are driven primarily by single retinal ganglion cells (RGCs). However, an LGN cell responds typically to less than half of the spikes it receives from the RGC that drives it, and without retinal drive the LGN is silent (Kaplan and Shapley, 1984). Recent studies, which used stimuli restricted to the receptive field (RF) center, show that despite the great loss of spikes, more than half of the information carried by the RGC discharge is typically preserved in the LGN discharge (Sincich et al., 2009), suggesting that the retinal spikes that are deleted by the LGN carry less information than those that are transmitted to the cortex. To determine how LGN relay neurons decide which retinal spikes to respond to, we recorded extracellularly from the cat LGN relay cell spikes together with the slow synaptic (‘S’) potentials that signal the firing of retinal spikes. We investigated the influence of the inhibitory surround of the LGN RF by stimulating the eyes with spots of various sizes, the largest of which covered the center and surround of the LGN relay cell's RF. We found that for stimuli that activated mostly the RF center, each LGN spike delivered more information than the retinal spike, but this difference was reduced as stimulus size increased to cover the RF surround. To evaluate the optimality of the LGN editing of retinal spikes, we created artificial spike trains from the retinal ones by various deletion schemes. We found that single LGN cells transmitted less information than an optimal detector could.