Two experiments, one using a between-Ss and one a within-Sa design, showed that response latencies to single attribute probe stimuli were longer when the target stimulus embodied two attributes (form and color) rather than a single attribute. The magnitude of this "mixed attribute effect" was influenced by the probability of attribute relevance, but the "probability effect" was mostly due to a "repetition effect," such that latencies on trials involving repetitions of the same attribute were shorter than when the relevant attribute was shifted. Implications of these results are drawn for the issues of holistic vs attributized representation of the target stimulus and serial vs parallel search of a set of attributes. Although no class of models can be ruled out on the basis of these experiments, constraints can be imposed on the versions of each that are compatible with the data presented.This report is concerned with the ways in which humans represent simple, visually presented stimuli in short-term memory and subsequently compare the representations with information from additional stimuli to guide a judgmental response. We introduced a novel variation on familiar experimental situations by presenting a multidimensional target stimulus, followed by unidimensional probes. We examine our data in relation to three broad classes of models: (1) "extraction models" in which the initial stimuli are represented in a holistic fashion with information extracted later according to the needs of the specific task (Lockhead, 1972); (2) parallel models, in which attributized representations are interrogated simultaneously over all attributes; and (3) serial models, in which attributes are evaluated one after another. Before considering the prior state of our knowledge regarding these matters, certain methodological comments seem appropriate.Two recent important papers by Townsend (1971Townsend ( , 1972 are sometimes cited as demonstrating the impossibility 'of distinguishing unequivocally between parallel and serial models of information processing. Townsend made no such claim. He did demonstrate that certain classes of parallel models could not be distinguished from certain other serial or hybrid models purely on the basis of the distributions of reaction times, although he gave examples of situations in which one of the formally equivalent alternatives could be rejected as psychologically implausible. A hoary tradition in psychology calls for the performing of "critical experiments": Two opposing theories are specified in considerable detail; one theory is falsified and the other is then claimed to have gained support. Townsend's papers call our attention to the fact that there are multiple alternative theories or models that could be proposed, and that individual experiments can at best