Factual passages about famous American presidents were read by over 500 subjects in two experiments. After an interval of 1 wk., 1 mo., or 2 mo., retention was tested for each of four types of information: main character, paragraph theme, sentence gist, and sentence wording. These vary in generality or abstraction, from the most abstract (true of all that the subject read) to the most specific details. Forced-choice tests were used that assessed retention of information at just one level, and that controlled for the reconstructive use of nonpassage knowledge. As expected, a hierarchy of retention was obtained:The more abstract the information, the better it was remembered at each of the three intervals tested. However, no evidence was found for differential forgetting rates. When memory codes are assessed individually, as in these experiments, their forgetting functions are found to be parallel. A second form of recognition test showed that thematic false recognitions increased over time, indicating that subjects do use related nonpassage information in reconstruction when the test permits this.