As a step toward the study of invention, an investigation dealt not with public arguments or the results of invention, but with arguments that may have occurred to the rhetor but were discarded. To avoid problems of self-presentation and retrospection, thinking aloud and reconstructive protocols were avoided in favor of providing 37 college students with a list of arguments they might have generated, and asking them why some of those arguments would be inappropriate. The Marwell and Schmitt list was chosen for the study because (1) it is reasonably detailed, (2) it is not strange in that the exemplars are straightforward, (3) the procedure allows subjects to select or reject as many arguments as they wish, and (4) the focus is the editing experience of going through the list, not the process of actually generating arguments. By coding subject responses according to a set of categories in roughly hierarchieal order, it was found that subjects edited on the basis of estimates of effectiveness, bias against certain kinds of arguments, self and interpersonal concerns, and a recognition that competent discourse must be true and relevant. The study concludes that the experiment design is workable, the coding system has reasonable validity, and the interesting results warrant further testing in more elaborate investigations. (A copy of the coding system is appended.) (CRH)