Although biomarkers are widely recommended, HIV prevention trials still assess participants' behavior primarily through self-report. We evaluate logical contradictions in self-report in similar randomized experiments evaluating HIV prevention interventions in women ages 15-25 followed for 1-3 years. We assess recanting for 7 behaviors in the 3 samples: reporting having “ever” engaged in a behavior and later reporting having never engaged in that behavior. Compared with two samples of clinic-recruited adolescents ages 15-20, we find substantially greater recanting in the community-recruited study of women ages 18-25. Using matching to identify overlap across the three samples, we evaluated whether recanting is associated with demographics, socioeconomic status, or behavior. Previous studies have found recanting associated with meaningful outcomes, but our preliminary analysis finds no meaning to this substantial recanting. Recanting may be due to non-overlapping characteristics of the samples, such as clinic vs. community recruitment.