This cluster randomised trial in KwaZulu-Natal South Africa, evaluated the implementation of a Feeding Buddies (FB) programme to improve exclusive breastfeeding (EBF) amongst human immunodeficiency virus infected mothers. Eight clinics were randomly allocated to intervention and control arms respectively. Pregnant women attending the prevention of mother-to-child transmission program and intending to EBF were enrolled: control (n = 326), intervention (n = 299). Intervention mothers selected FBs to support them and they were trained together (four sessions). Interviews of mothers occurred prenatally and at post-natal visits (day 3, weeks 6, 14 and 22). Breastfeeding results were analysed (Stata) as interval-censored time-to-event data, with up to four time intervals per mother. EBF rates at the final interview were similar for control and intervention groups: 44.68% (105/235) and 42.75% (109/255) respectively (p = 0.67). In Cox regression analysis better EBF rates were observed in mothers who received the appropriate training (p = 0.036), had a community care giver visit (p = 0.044), while controlling for other factors. Implementation realities reduced the potential effectiveness of the FBs.