Research on iconic memory is reviewed. Specific issues discussed include the duration of the icon, effects of stimulus variables, types of information lost, selection, processing capacity, and scanning. More general issues include the level of encoding in the icon and its relation to short-term memory. It is also argued that a number of experiments do not show what they were intended to show because of possible methodological problems. The view is developed that iconic memory is postretinal but uncoded; nor is it influenced directly by strategies or subsequent mechanisms.The idea of a brief, time-dependent memory serving as an early stage in the analysis of information has existed for a long time (MulIer & Pilzecher, 1901, cited in Woodworth & Schlosberg, 1954. Hebb (1949) employed the concept in his two-stage theory of memory. He suggested that memory consisted of a brief neural activity phase (lasting approximately 1/2 sec) and a second permanent, structural trace. In his theory, the function of the activity phase was to maintain the information until the structural celI assembly could be established. Subsequently, many other investigators have also incorporated a two-stage memory concept into their theoretical frameworks. For example, Broadbent (1958) used the concept to explain some of his observations in dichotic listening.The first clear behavioral evidence in support of such a time-dependent memory came in Sperling's (I 960) work, in which he showed a decline in accuracy during the first few hundred milliseconds following a brief tachistoscopic exposure (cf, Boynton, 1972). Sperling's work was soon supported by the results of Averbach and Coriell (1961), and subsequent investigators have generally found support for Sperling's results. Theorizing about Sperling's work came more slowly, but Neisser's (1967) theoretical discussions served to solidify the notion of a rapidly decaying memory. Of course, as the empirical work became available, the notion of iconic memory has become more definite. In general terms, iconic memory can be described as a large-capacity, short-duration image. It is a central memory and appears to hold material in a fairly literal form. Obviously, the mechanism is sensory-specific, and thus theoretical notions, including iconic memory, are limited to cases less general than Hebb's suggestions concerning a brief activity trace. Nevertheless, some theorists appear to assign iconic memory a role not unlike that required by Hebb (e.g., Haber, 1971).1In addition, there are practical implications of the phenomenon as well. For example, in everyday tasks, such as reading, we know that information is primarily taken in during fixations (Erdmann & Dodge, 1898, cited in Woodworth & Schlosberg, 1954Latour, 1962;Yarbus, 1967). Eye movements are important, of course, in determining the rate of reading and the sequence of fixations. Clearly, however, mechanisms other than eye movements must be involved in making spatial-to-temporal conversions on information taken in during a single fixation. [F...

A typkal trial of this masking experiment in\·olves. in quick succession. presentation of five letters. evocation of an eye mowment. and presentation of a spatially localized mask. either a visual-pattern mask or a metacontrast ring. The effect of the mask is to suppress the report of the letter that stimulates the same retinal location, even though the mask appears to cO\'er or surround the letter whose position in real space it shares. "-tasking is. however. weaker when the eyes move than when they do not. An auxiliary experiment suggests that the spatial aspects of observable (reportable by S) stimulus persistence are unaffected by eye movements. and therefore that observable persistence differs from that susceptible to masking.

In eleven hemispheres of nine marmoset monkeys (Callithrix jacchus), we have investigated the thalamo-cortical organization of the projections from the pulvinar to the striate and prestriate cortex. In each experiment, single or multiple injections of various retrograde fluorescent tracers were injected into adjacent regions or areas. In two experiments, horseradish peroxidase (HRP) was injected into the lateral geniculate nucleus (LGN) and the lateral pulvinar, respectively. The results show that the thalamo-cortical projection from LGN to striate cortex and from pulvinar to the prestriate cortex are similarly organized, but the geniculo-striate projection is more precise than the pulvinar-prestriate projection. The pulvinar-prestriate projection is topographically organized and preserves topological neighbourhood relations. Projection zones to the various visual areas are concentrically wrapped around each other. The projection zone to area 18 constitutes a central core region. It begins ventro-laterally in PuL where the pulvinar is in contact with the LGN. This contact zone we called the hilus region of the pulvinar. The area 18-projection zone stretches as a central cone into the posterior pulvinar through PuL and into PuM. It is surrounded by the projection zone to the posterior belt of area 19 and this in turn is surrounded by the projection zone to the anterior belt of area 19. The projection zones to area 19 are then surrounded medially and dorsally by zones projection to the temporal and parietal association cortex, respectively. The projection zone to area MT is located medio-ventrally in the posterior pulvinar (PuIP and surrounding nuclei) and coincides with a densely myelinated region. Area 17 also receives input from the pulvinar but probably predominantly in the region of the central visual field. The pulvinar zone projecting to area 17 is located ventrolaterally from the central core region projecting to area 18 and is contiguous laterally with the LGN. If the positions of the vertical and the horizontal meridian in the pulvinar correspond to those in the respective cortical projection zones, a second order visual field representation such as found in area 18, with the horizontal meridian split at an eccentricity of about 7-10 degrees, can also be recognized in the pulvinar.

A reaction-time experiment was carried out to examine the relationship between naming and categorization. Ss were shown one item at a time and asked either to name the item or to categorize the item as a letter or number. The size of the stimulus set was varied systematically across Ss.Naming time increased as the stimulus set size increased; categorization time could be predicted by the time required for naming, plus a constant. These results were interpreted as indicating that naming must precede categorization.In a previous experiment studying iconic memory, Dick (1969) tachistoscopically exposed eight-item alphanumeric displays; this display was followed by a postexposure coded auditory cue which indicated to the S how to report the display. Three groups of Ss were shown a common set of visual stimuli; one group reported according to the spatial aspects of the display-the top row, the bottom row, or both rows; a second group reported according to color-red items, black items, or both; the third group reported according to category of the items-letters, numbers, or both. The auditory cue was systematically varied with respect to the stimulus exposure. Results of this experiment indicate that, as the report cue was delayed postexposurally, accuracy decreased for both the color-and spatial-report groups, but did not decrease for the c1ass-or category-report group. Thus, there was evidence for loss of spatial and color information but not for category information as a function of delay of the report cue.These results were interpreted in the following way. Dick (1969) argued that two stages of perceptual processing or memory are necessary to account for the results. The first stage consists of iconic memory (Neisser, 1967) or a sensory register (Atkinson & Schiffrin, 1968). The second stage consists of short-term storage, which is much more verbal in nature than is the iconic stage. Color and location represent physical dimensions of the stimulus and, therefore, these are analyzed in the sensory register. Spatial position, color, and category information were represented in terms of their names since all Ss reported the stimuli by naming the items that were in the display. In order to

Perception is viewed as a process in which attributes of a stimulus are analyzed in step-wise fashion. Two experiments were carried out in which the attributes of spatial location and identity were examined for two types of materials. The results indicate that identification of letters requires more stimulus energy than identification of lines. In turn, there were no differences between identification of lines, localization of lines or localization of letters. In general, these results support a hierarchical processing hypothesis.RECENTLY, SEVERAL INVESTIGATORS have suggested that perceptual processing may be composed of a series of hierarchical tests or analyses performed on a sensory input (Egeth, 1967;Posner & Mitchell, 1967). This suggestion is not particularly new since it has been known for a long time that brightness and form recognition thresholds differ considerably (Forgus, 1966), apparently because analysis of form occurs corticaUy (e.g., Lashley & Franks, 1934), while certain aspects of brightness are processed at a lower level. What is interesting, however, is the suggestion that within form recognition it is possible to make a much finer distinction in terms of the physiological level at which different types of information are analyzed. Some support for such a position has been provided physiologically by Hubel and Wiesel (1959;Hubel, 1963). These investigators have shown that a variety of different neurons exist in the striate cortex of the cat; each of these neurons responds to a specialized type of stimulation. "Simple" cells respond only to lines or bars of particular orientations with each cell responding to its own special orientation and physical location. For example, a cell with vertically oriented field will respond strongly only if die stimulus is oriented vertically in a particular area of the retina; when the stimulus is oriented in any other direction 1 The early stages of this research and the computer analyses of the data were supported in part by Defence Research Board of Canada, Grant No. 9401-26 to Dr. M. P. Bryden of the University of Waterloo. The other aspects of these experiments were supported by National Science Foundation Grant No. GB-7848. The authors would like to thank R. H. Jackson for assistance in testing the Ss in Experiment n and Dr. D. J. K. Mewhort for computer analysis of the data and for comments.2