Objective: To determine the feasibility, accuracy and cost-effectiveness of a rapid, on-site, HIV testing strategy in a rural hospital, and to assess its impact on test turnaround time and the proportion of patients post-test counselled. Design: Prospective comparison of two testing strategies |double rapid test on-site versus central enzyme-linked immunosorbent assay (ELISA)-based testing|, and an economic evaluation. Setting: Hlabisa Hospital, a rural South African district hospital. Patients: A total of 454 consecutive adult inpatients requiring and consenting to HIV testing as part of their clinical management. Main outcome measures: Concordance between rapid tests, and between the rapid and ELISA strategies, test turnaround time, proportion of patients post-test counselled, and cost-effectiveness. Results: HIV seroprevalence was 49.6%. Both rapid tests were concordant in all patients |one-sided 95% confidence interval (CI) of probability, 99.3-100|. The rapid strategy was 100% sensitive (95%, CI, 97.9-100) and 99.6% specific (95%, CI, 97.2-100) compared with the ELISA strategy. The mean interval between ordering a test and post-test counselling fell from 21 days prior to the introduction of the rapid test strategy to 4.6 days after its introduction (P < 0.00001). The proportion of patients post-test counselled increased to 96%, from 17% after the introduction of the rapid test strategy (P< 0.00001). By using a double rapid test strategy the cost per patient post-test counselled was almost halved to US$ 11. Accuracy of the rapid strategy was not substantially increased by performing two tests. Conclusion: In high prevalence, resource-poor settings, rapid, on-site HIV testing is feasible, accurate and highly cost-effective, substantially increasing the number of patients post-test counselled. A single rapid test may be sufficient.