The impact that malnutrition at the time of starting antiretroviral therapy (ART) has on survival and the CD4 count response is not known. A retrospective cohort study of patients attending the national HIV referral centre in Singapore who had a CD4 count less than 250 cells/microL and a measurement of body weight performed at the time of starting ART was carried out. Demographic and clinical variables were extracted from an existing database. Body mass index (BMI) was calculated from the weight in kilograms divided by the square of the height in metres. Moderate to severe malnutrition was defined as BMI less than 17 kg/m(2). Intent-to-treat Cox models were used to determine the predictors of survival. A total of 394 patients were included in the analysis, of whom 79 died during a median study follow-up of 2.4 years. Moderate to severe malnutrition was present in 16% of patients at the time of starting ART, and was found to be a significant independent predictor of death [hazard ratio (HR) 2.19, 95% confidence interval (CI) 1.29-3.73, P=0.004 for those with BMI<17 compared with those with BMI>18.5] as were stage of disease (HR 2.47, 95% CI 1.20-5.07, P=0.014 for those who were at stage C compared with those at stage A) and the type of ART [HR 0.50, 95% CI 0.27-0.93, P=0.03 for highly active antiretroviral therapy (HAART) compared with non-HAART treatment]. Malnutrition did not impair the magnitude of the increase in CD4 count at 6 or 12 months. Malnutrition at the time of starting ART was significantly associated with decreased survival, but the effect appeared not to be mediated by impaired immune reconstitution. Given the increasing access to ART in developing countries and the high frequency of HIV-associated wasting, studies of nutritional therapy as an adjunct to the initiation of HAART are urgently needed.