Researchers have proposed that graphical efficacy may be determined, in part, by the nature of the perceptual interactions that exist between attributes used to create graphical displays, One extreme type of interaction is integrality, in which two or more physical dimensions are represented as a single psychological dimension in the observer. An alternative type of interaction is configurality, in which a global emergent dimension is available to the observer in addition to the component attributes, Thirteen stimulus sets, each composed of attributes commonly used in the design of graphs, were submitted to the performance-based diagnostics of integrality and configurality. Analyses suggest a continuum of configurality among the present stimulus sets, with little evidence for integral graphical attributes. The configural pattern of results was more common when two identical dimensions were paired (homogeneous stimuli) than when two different dimensions were paired (heterogeneous stimuli). However, there was no evidence that pairs of dimensions belonging to a single object (object integration) were any more configural than dimensions belonging to different objects. Object integration was, however, consistently related to inefficient performance in tasks requiring the filtering of one of two component dimensions.How may geometric properties and color be most effectively employed to communicate quantitative information? What, in short, makes a good graphical display? These questions represent the research agenda for what DeSanctis (1984) calls comparative graphics. One approach to comparative graphics has been to systematically study the accuracy with which we can identify and discriminate stimulus levels along continuous physical dimensions, and this information is routinely presented in introductory discussions of display design (e.g., Hutchingson, 1981;McCormick & Sanders, 1982). However, this tradition, which emphasizes the coding of individual quantitative variables as individual physical dimensions, has recently been supplemented by interest in the perceptual interactions that may occur when combinations of such dimensions are used to represent multiple variables. In the present paper, we consider the utility of two notions of attribute interaction-dimensional integrality and dimensional configurality-for the prediction of graphical efficacy.