The role of the response probability convention in reporting results from the 1992 National Adult Literacy Survey is explored, using interviews with more than 26,000 adults and young adults. In order to summarize what respondents of a particular proficiency can do, it is convenient to adopt a convention for a sufficient response probability that is stringent enough to ensure that people at a lower bound can do what the task requires most of the time. Sections of the paper focus on the measurement of literacy in this survey, sources of the data, and item response theory scaling methods and item characteristic curves. The use of item mapping to anchor the prose literacy scale by locating specific tasks along it, using a response probability convention, and the literacy levels created for the survey in order to generalize beyond specific tasks to the more abstract abilities underlying the scales are also discussed. Other sections consider the relationship of the response probability convention to the cut points between the literacy levels and the variation in the proportions of the adult population reported to be in each prose literacy level as a function of the response probability convention. Results indicate that if the adult literacy program were to adopt the same response probability convention as that used in the National Assessment of Educational Progress, the proportion of the population in literacy Levels 1 and 2 would drop by 15% and the proportion in literacy Level 5 would increase by 9 percentage points. (Contains 8 figures and 10 tables.) (SLD)