OBJECTIVE: To investigate determinants of breastmilk RNA viral load among HIV-infected South African women, with particular attention to infant feeding mode and subclinical mastitis. DESIGN: Observational, longitudinal study. METHODS: Information on current infant feeding practice and a spot milk sample from each breast were obtained from 145 HIV-infected lactating women at 1, 6 and 14 weeks postpartum. The sodium/potassium (Na+/K+) ratio in milk was taken as an indicator of subclinical mastitis. The association between milk RNA viral load and maternal and infant characteristics was investigated using uni- and multivariate models. RESULTS: Milk viral load was below the limit of detection of the HIV RNA assay (< 200 copies/ml) in 63/185 (34.1%), 73/193 (37.8%) and 68/160 (42.5%) of samples at 1, 6 and 14 weeks, respectively. Multivariate models predicted between 13 and 26% of variability in milk viral load in the first 14 weeks. Low blood CD4 cell count (< 200 x 10(6) cells/l) during pregnancy and raised milk Na+/K+ ratio were significantly associated with raised milk RNA viral load at all times, but there were no consistent associations between infant feeding mode and RNA viral load in milk. There was a non-significant trend for the six infants known to be infected postnatally, compared with the 88 infants who remained uninfected, to have been exposed to breastmilk of higher viral load at each time point. CONCLUSIONS: Breast milk HIV RNA viral load in the first 14 weeks of life varied; high levels were associated with subclinical mastitis and severe maternal immunosuppression. Multivariate models had limited predictive value for milk RNA viral load, illustrating the multiple contributors to viral load.