The third edition of the "Boston Diagnostic Aphasia Examination" (Goodglass, Kaplan, and Barresi) introduced standardized procedures for coding discourse samples elicited using the well known Cookie Theft illustration. To evaluate the reliability of this discourse coding procedure, a transcribed sample was coded by 14 novice examiners using the Goodglass et al. coding criteria. Percentage of agreement scores between novice and expert coders were poor (with an average of less than 50% agreement). This paper describes factors that may have contributed to the poor interjudge agreement, and one approach to improving agreement is described. (Contains 1 table and 2 figures.)