The accuracy with which graphical elements are judged was assessed in a psychophysical task that parallels the real-life use of graphs. The task is a variant of the Metfessel-Comrey constant-sum method, and an associated model based on Stevens's law is proposed. The stimuli were horizontal and vertical lines, bars, pie and disk slices, cylinders, boxes, and table entries (numbers). Stevens's law exponents were near unity for numbers and 1-dimensional elements but were also close to 1 for elements possessing 2 or 3 apparent dimensions--subjects accommodate extraneous dimensions that do not carry variation, changing the effective dimensionality of the stimulus. Judgment errors were small, with numbers yielding the best performance; elements such as bars and pie slices were judged almost as accurately; disk elements were judged least accurately, but the magnitude of the errors was not large.