We begin this report by reviewing some deficiencies in the traditional view of "iconic memory," and outlining some current questions about the initial representation and processing of alphanumeric arrays. These questions invite the application of new experimental methods, one of which is developed in the present report. The report contains informal accounts of five experiments concerned with visual memory for small arrays of characters, and with the transformation of this memory during the first second after display offset. The central hypothesis, which was supported, is that an initial "spatial" representation is distinguished from a "list memory" into which it is transformed by being directionally symmetric. That is, there exists a scanning mechanism that can operate on the spatial representation (but not on the list memory) either left-to-right (forward) or right-to-left (backward) so as to produce performances that are qualitatively the same and quantitatively similar. The main procedure is a reaction-time variant of the "whole report" method (Sperling, 1960), applicable in the study of accurate performance based on small arrays. Subjects recited the names of all the digits in a briefly-presented array of from one to five digits, either forward or backward. They were instructed to complete the response as soon as possible after a tone-burst probe. We studied performance as a function of probe delay when reciting direction was specified before display onset, and also when it was uncertain and specified only by the probe itself. The principal measure was response latency, which we found to be strongly influenced by array size, probe delay, and (under some conditions) directional uncertainty.

With directional certainty (Experiment 1) an initially large effect of array size on latency declined with probe delay, indicating a transformation of visual memory that was essentially complete in less than 1 sec. The time course of the transformation closely resembles that of a transformation investigated in an earlier series of studies on visual memory (Sternberg, Knoll, & Leuin, 1975) in which we employed a reaction-time variant of a "partial-report" procedure and measured changes in search rate with probe delay; the two transformations may be the same. With directional uncertainty (Experiment 2), the failure of the array-size effect to decline with probe delay indicated that the transformation was at least partially inhibited; other results suggested that even after 1 sec, when reciting was required in the unexpected direction it was based on a persisting spatial representation rather than a list memory. In Experiment 3 we showed that shifting of attention to the appropriate "end" of the internal array-representation played at most a minor role in the continuing large effect of array size on latency with directional uncertainty. In Experiment 4 we distorted the displayed characters to slow their identification. With directional certainty the effect on latency disappeared as the probe was delayed, consistent with transformation into a nonvisual form. With directional uncertainty, however, the distortion effect persisted, consistent with the use of a visual memory, even after delay. In Experiment 5 we attempted to prevent subjects from using a spatial representation of the digit string by presenting digits sequentially in one location. Although forward-reciting after a 2-second delay with directional certainty was similar to performance based on an array, performance with directional uncertainty was radically inferior. The substantial advantage of array presentation shows that even after a delay of as much as 2 sec, the internal representation produced by an array (but not by a sequence) can serve as a basis for efficient performance.