The verbal descriptions varied in terms of the consent given by the female and the degree of force or coercion used by the male. Each subject was tested in two laboratory sessions, and rape indices were calculated as the ratio of rape arousal to consenting arousal. The discrimination between consenting and nonconsenting cues was stronger in nonrapists than rapists. This difference was observed mainly because of an increase in the nonrapists' discrimination in the second session, in which nonrapists' arousal to consenting cues was enhanced. Based on the whole sample, test-retest reliability coefficients were unacceptably low. Eliminating subjects exhibiting only minimal arousal according to various criteria yielded higher coefficients in both groups. However, in the nonrapist group, acceptable levels of reliability were only reached after eliminating all subjects who did not reach at least 75 percent of full erection (over half the sample). We concluded that the reliability of the rape index is questionable, particularly when there is doubt as to whether the subject being tested is a rapist or nonrapist, or when testing only nonrapists. 1 table, 1 note, 30 references. (Author abstract)