We compared 202 households with an HIV-infected member, and 202 unaffected neighbouring households, in one rural and one urban area in Free State province, South Africa. Interviews were conducted with a key informant in each household, at baseline and 6, 12 and 18 months later, with 87% follow-up. Results of the baseline and 6 months study were published previously; this paper reports on trends over the full 18 months of follow-up. We analysed changes in income and expenditure and their relationships with morbidity and mortality using multiple regression. Over 2 years, a third of affected households reported at least one death, mainly due to infectious disease. Affected households’ real expenditures decreased over time, while real incomes did not, absolutely and relative to unaffected households. Mortality and morbidity independently predicted lower expenditure (adjusted relative expenditure 0.86 and 0.84, respectively), mortality predicted lower income (adjusted relative income 0.80) and higher expenditure predicted less morbidity and mortality (adjusted odds ratios for a US$100 monthly difference 0.78 and 0.65). Households’ economies and HIV/AIDS were thus dynamically related, and occurred against a background of progressive poverty. Expenditure was a more sensitive indicator of the economic impact of HIV/AIDS than was income.