The sensitivity of naturalistic prose material to retroactive interference (RI) was investigated in three experiments, with due consideration given to the conditions known to produce RI in nonprose material. In experiment 1, unrelated, related, and nearly identical (parapharase) prose passage conditions yielded a pattern of results very similar to those obtained with non-prose material, with only the related condition associated with significant amounts of RI. Based upon these results and a few simple assumptions, a classificatory scheme was developed within which predictions were made regarding RI in related prose material. Several such predictions were tested and confirmed in Experiments 2 and 3. It was concluded that the many previous failures to detect RI in the retention of prose may be attributed to factors other than any inherent insensitivity of the material itself, and that RI may be just as predictable in prose as it is non-prose material.For many years retroactive interference (RI), said to have occurred when memory for previously acquired information is reduced by subsequent learning, was a dominant theme in the field of human learning and memory. Indeed, RI is one of the most frequently studied phenomena within all of experimental psychology. Hence the basic paradigm is fairly well known to students of psychology. In the prototypical experiment, experimental groups are presented with two experimental lists (usually paired-associates) and then, following a retention interval, they are tested for their memory of the first list. The amount of RI obtained is determined by comparing the retention test scores of the experimental groups to the scores of a control group, that either did not receive a second list or learned a totally unrelated list. In most studies the control group retains significantly more of the first list than the experimental groups; an effect that defines the phenomenon of RI.One of the most important single factors governing the occurrence and magnitude of RI is the relationship among the elements of the learning material in the lists. Many studies, for example, have shown that increased formal (e.g. from GAW-634 to GAW-COV) or semantic (e.g. from elated-left to elated-high) similarity of the material, provided it is not too closely related, tends to increase the degree of interference (Bugelski and Cadwallader, 1956;Friedman and Reynolds, 1967; J. A. McGeoch and G. 0. McGeoch, 1937;Osgood, 1946;Postman, Keppel and Stark, 1965;Postman and Stark, 1969; Underwood and Richardson, 1956). In addition, semantic relationships other than simple similarity of meaning have been found to be effective in producing RI. For example, meanings that share the same category membership also produce heavy retroactive losses (Birnbaum, 1974;Shuell, 1968). In these studies, significantly