Pooling specimens when testing them in large numbers can save scarce resources and several earlier reports have indicated this to be a feasible strategy. In an HIV antibody mass screening test carried out in our laboratory, we used Dorfman's two-stage model. We sought to establish the optimal number of specimens in a pool, and to achieve maximum efficiency while maintaining both sensitivity and specificity. Before testing for HIV antibody, five positive samples were placed in a set of 1012 sera in a double blind manner, one positive sample into a second set of 1012 sera and none in a third set. The positive rate was assumed to be 0.2% for each set of 1012 sera. As indicated by our model, 22 individual serum samples were placed into each of 46 pools which, when tested by particle agglutination assays, lead to the identification of all positive samples. We concluded that the prevalence rate can be estimated in the first stage, 95% confidence intervals were given, and the efficiency rate could be calculated for the identification of all infected specimens in a large number of samples showing low prevalence rates.