This paper reexamines the visual search process, and visual information processing more generally, from a perspective of the continuous flow of information and responses through the visual system. The results from three experiments are reported which support the continuous flow conception: Information accumulates gradually in the visual system, with concurrent priming of responses. The first two experiments investigated the processing of display stimuli which varied in size and figure-ground contrast in a nonsearch task, and provided evidence confirming a continuous flow model. Experiment 3 employed an asynchronous onset of target and noise and provided convergent evidence of the accumulative nature of information and response priming in visual processing.

Much recent visual information processing research has employed linear or rectangular letter displays subtending up to 8 deg of visual angle in width. Subsequent theory based on these data has assumed the simultaneous availability to some central processor of all n targets. Data on retinal acuity suggest that this assumption is gratuitous. Six subjects served in a tachistoscopic speeded letter recognition task wherein display luminance, retinal target locus, number of alternative target positions, and stimulus degradation by superimposition of a dot grid were manipulated orthogonally. Identification reaction times increased on the order of 100 msec as targets were located 3 deg from the fovea center, and results under convergent operations in the design suggested that these reaction time increases were a manifestation of peripheral stimulus degradation factors (such as acuity, masking, and luminance) rather than more central cognitive processes.

Much recent research in visual information processing has employed a methodology resting on the assumption that a noise mask following presentation of a target stimulus terminates processing of that target. In the absence of appropriate controls, such a methodology is viable only insofar as an erasure theory of masking is valid. However, the phenomena from which the erasure position has derived its strongest support have been subject to alternative theoretical explanations, the most general of which is that of temporal integration. The experiment reported here tested these alternatives. Twelve subjects served in a tachistoscopic study designed to determine whether the same noise field of dots could either erase a degraded target digit or facilitate target identification through temporal integration, under both forward and backward masking paradigms. This was found to be the case, and the results were interpreted as consistent with an integration theory of masking and as incompatible with an erasure conception. The results suggested that efforts to control target processing time through display of a visual noise pattern subsequent to target presentation are methodologically inadequate when devoid of some basic control operations.

Ten subjects served in a speeded-recognition task wherein letter targets were varied from .143 deg to 2.14 deg of visual angle in height. Identification latencies were found to decrease on the order of 40 msec for both blocked and randomized presentations of sizes as target size increased. Beyond a size of approximately .75 deg of visual angle, reaction time became asymptotic, and blocked presentation of target size was judged to be faster than randomized. These results were interpreted to support the view that a percept in the visual system develops gradually over time, permitting gross figure-ground differentiation very early in processing, but discrimination of fine detail only later. Implications of these results for human-factors applications are discussed.

A FORTRAN program to compute latency-operating characteristics (LOCs) is presented. Latency and accuracy scores from within-subject designs having equal cell frequencies and containing no missing data are appropriate. Output includes both latency measures (mean and median reaction times) and accuracy measures (d', d 12 • log odds. and log eta) for each latency band. The user determines the number of latency levels into which the data are partitioned. Input, output, and program subroutines are discussed.