BACKGROUND: It has been suggested that a new strategy for HIV prevention, 'Universal Test and Treat', whereby everyone is tested for HIV once a year and treated immediately with antiretroviral therapy (ART) if they are infected, could 'eliminate' the epidemic and reduce ART costs in the long term. METHODS: We investigated the impact of test-and-treat interventions under a variety of assumptions about the epidemic using a deterministic mathematical model. RESULTS: Our model shows that such an intervention can substantially reduce HIV transmission, but that impact depends crucially on the epidemiological context; in some situations, less aggressive interventions achieve the same results, whereas in others, the proposed intervention reduces HIV by much less. It follows that testing every year and treating immediately is not necessarily the most cost-efficient strategy. We also show that a test-and-treat intervention that does not reach full implementation or coverage could, perversely, increase long-term ART costs. CONCLUSION: Interventions that prevent new infections through ART scale-up may hold substantial promise. However, as plans move forward, careful consideration should be given to the nature of the epidemic and the potential for perverse outcomes.