Little is known about the ability of women to adhere to recommended feeding strategies to prevent mother-to-child HIV transmission (MTCT) from breast milk. We conducted a pilot study in rural Botswana to prevent MTCT from breast milk. Women were randomized to formula feed their infants or to exclusively breastfeed while providing prophylactic zidovudine. Women who chose to formula feed independently were also followed. Among those with > or = 3 postpartum visits, none of 31 women assigned to breastfeed did so exclusively for 5 months. Seven (22%) of 32 women in the formula arm definitely or probably breastfed by self-report or as witnessed in maternity, and evidence of breast milk on physical examination was present in 50% of women in > or = 2 visits beyond 1 month. Three (18%) of 17 women choosing formula definitely or probably breastfed, and breast milk was present on exam in 53%. We conclude that adherence to 5 months of exclusive breastfeeding did not occur, and that adherence to exclusive formula feeding was sub-optimal and potentially over reported. Breast examination may be a useful adjunct to self-report, but needs to be validated and standardized. Low adherence to infant feeding strategies that differ from local norms will reduce their effectiveness in preventing MTCT.