In the CAPRISA 004 trial, adherence was estimated as the proportion of reported sex acts covered by two gel doses, which was assessed by counting returned empty gel applicators. The returned empty applicators were inspected visually in a standardized manner for residue on the outside of the applicator, as an indicator of vaginal insertion. Over 15 months, spanning 11,839 study visits by 838 women, a total of 59,800 returned empty applicators were inspected. By visual assessment, 77.5 % of these applicators appeared to have been inserted. To test the accuracy of the assessment we fitted a Cox model and found that the risk for HIV infection was doubled when less than half of the returned empty applicators had been assessed as not inserted in the vagina. Visual inspection enhanced both the accuracy of the adherence measurement and aided identification of mechanical problems with applicator use experienced by women in the trial.