AbstractEpidemic models are developed to evaluate the potential benefits of voluntary confidential testing (VCT) for HIV. The number of tests required to prevent one case (called the economic ratio ER), and the number of cases prevented by the screening programme are calculated. Methods differ from commonly used epidemic models because susceptibles and infecteds are subdivided according to testing status, to permit modelling of the efficacy of knowledge of HIV status in retarding epidemic spread. The methods also allow for subdivision into several homogeneous subpopulations whose members may be attracted in varying degrees to members of other subpopulations. Thus the common assumption of free mixing across subpopulations is relaxed. The following conclusions which refer to a population of 100,000 persons over a period of 5 to 15 years, are robust to wide variations in assumed parameter values and other aspects of modelling:1VCT prevents hundreds or thousands of infections in isolated high risk populations and ER values are typically less than 100, making VCT very attractive economically.2VCT prevents only a few infections in isolated low risk populations with initial prevalence 0‐1 per cent or less, and the ER values are well above 2000. However, in ‘low risk’ populations with 1 per cent initial prevalence, tens or hundreds of infections may be prevented, and ER values fall below 2000 for plausible spread rates, indicating that VCT may be economically feasible in such settings.3In a mixed population of gays, bisexuals and heterosexuals, a VCT programme that aims primarily at the gay/bisexual subpopulations prevents more disease in the gay/bisexual populations, prevents more disease in the heterosexual populations, and requires fewer tests per case prevented than a VCT programme that tests all subpopulations eq