The Evidence for Contraceptive Options and HIV Outcomes (ECHO) trial found no substantial difference in HIV acquisition risk between women randomised to injectable intramuscular depot medroxyprogesterone acetate (DMPA-IM), copper intrauterine device (Cu-IUD) or the levonorgestrel (LNG) implant. We evaluated post-randomization sexual behavior using an objective marker of condomless vaginal sex in a subset of participants. We conducted a sub-study among 458 ECHO participants at three sites (Cape Town, Johannesburg, Kisumu) to evaluate the frequency of condomless vaginal sex, measured by prostate specific antigen (PSA) detection in vaginal swabs, collected at the month 6 and final visit and the concordance of self-reported condomless vaginal sex with PSA detection, by randomized arm. We compared PSA detection frequency and concordance of PSA and self-reported condomless vaginal sex, by randomized group using Cochran–Mantel–Haenszel tests and adjusted generalized logistic growth curve models. PSA was detected less frequently in the DMPA-IM (16%), compared to the Cu-IUD (21%) and LNG implant (24%) groups, although results were not statistically significant in the unadjusted model when accounting for pre-specified multiple-testing criteria. There were significant differences in PSA detection between the DMPA-IM and LNG-implant groups (odds ratio 0.61 (95% CI 0.40, 0.94) in the adjusted model. There was moderate discordance between self-reported condomless vaginal sex and detection of PSA that was similar across randomized groups. These data suggest that women randomized to Cu-IUD and LNG implant may have had condomless sex more frequently than women randomized to DMPA-IM. The discordance between detectable PSA and self-reported sexual behaviour has important implications for design of future HIV prevention studies.