Short courses of antiretroviral drugs have greatly enhanced the prospect of reducing mother-to-child HIV transmission. Yet transmission by breast feeding clouds hopes for this seemingly simple intervention. We revisit mathematical models to assess the competing risks associated with feeding by breast vs. formula. These indicate that, in the less developed world where the HIV epidemic predominates, neither option, unmodified, offers a reasonable choice for HIV-positive women. Where infant mortality rates are greater than about 40 per 1000 live births, if formula were made available to HIV-infected women only, the excess number of deaths that would result from formula use would be approximately the same or greater than the number of HIV infections that might be prevented. Only at lower infant mortality rates, less than about 40 per 1000, is the risk greater on the breast. There are thus no good grounds for the total avoidance of breast feeding under all conditions. Research to develop and test safer infant feeding alternatives is an urgent priority. On the one hand, ways to reduce HIV transmission while preserving breast feeding, as exclusive breast feeding could do, need to be fully tested. On the other hand, ways to reduce non-HIV morbidity and mortality associated with formula feeding, as educational or sanitary interventions could do, equally need testing. With either approach, a necessary foundation for implementing all the core components of preventing mother-to-child HIV transmission is competent counselling for mothers. Innovative approaches are needed to mobilise and train effective counsellors among health care workers and, as appropriate, community members.