Loftus's (1993bLoftus's ( , 1995 recommendation that investigators routinely include plots of appropriate means along with 95%confidence intervals or some other indication of variability has considerable merit, yet we agree with Morrison and Weaver (1995) that such plots can supplement but not supplant the usual reporting of analysis of variance results. Providing them may be easier than Loftus and Masson (1994) indicated, especially when error bars are understood as supplemental descriptive devices. We suggest a general, unified approach that applies to the explication of both between-and within-subjects effects. Variability is estimated separately for each group of scores identified as different by analysis because this serves description better. Raw scores are used for between-subjects effects, scores adjusted for between-subjects variability for within-subjects effects. All computations and figures are easily effected using common spreadsheet programs.In a recent series ofarticles in this journal and elsewhere, Loftus (1991Loftus ( , 1993aLoftus ( , 1993b argued for replacing traditional analysis of variance (ANaYA) results with pictorial representations of those results based on error bars, which he referred to as a plot-plus-error-bar (PPE) approach. He argued that traditional hypothesis testing emphasizes a binary yes-it-is or no-it-isn 't decision (see also Cohen, 1990;Rosnow & Rosenthal, 1989) that obscures two critical aspects ofany study, the strength of the associations between variables (i.e., the magnitude of the effects) and the relations among parameters, whereas error bars render these important aspects visible.In response, Morrison and Weaver (1995) argued that presentation oferror bars may supplement, but cannot supplant, standard hypothesis-testing procedures. Although agreeing that error bars are often useful and desirable, Morrison and Weaver noted that when designs include repeated measures or within-subjects factors, the definition and computation of the appropriate standard error is unclear. Loftus (1995) agreed that whatever one does technically, insight into research questions remains paramount and, drawing on another recent paper (Loftus & Masson, 1994), proposed definitional and computational solutions to the technical issues Morrison and Weaver raised concerning within-subjects error bars.Error bars are of many kinds. Discussing Loftus's PPE approach in general, Morrison and Weaver (1995) asked, why the standard error? If the intent is to provide a descriptive measure of variability, why not simply use the standard deviation or some other descriptive measure?