The order of visual processing: Top-down, bottom-up, middle-out, or none of these?

To investigate whether global or local visual information is processed first, several researchers have used a large letter (global information) made up of smaller letters (local information) and measured speed of response and interference effects between global and local components. Kinchla and Wolfe (1979) argued that processing was neither invariably top-down (global-to-local) nor bottom-up (local-to-global). Instead, they argued that the fastest identification of letter shapes is given at an optimal visual angle of about 2° and proposed a middle-out processing model wherein elements of optimal size are processed first. In the present study, three-level figures were used in a direct test of that hypothesis, with subjects monitoring all three levels to detect target letters occurring at any level. The results are not compatible with any of the three orders of processing that have been proposed, and an interpretation in terms of the discriminability of the information available at each level is advocated. It is concluded that the two-level paradigm with letters is inadequate for investigations of the order of processing.