Objective To evaluate the use of saliva specimens for the detection of HIV antibodies among high-risk groups in epidemiological studies. Design Testing of saliva specimens collected by different methods from individuals with known HIV status. The most reliable method was examined for its usefulness in a field study among a high-risk group. Methods Saliva samples were obtained either by using a cotton-wool roll ('Salivette') or as 'whole saliva'. HIV antibodies were determined using commercial enzyme-linked immunosorbent assays (ELISA). Confirmation was performed using a line immunoassay or an immunoblot assay. Results In 'Salivette' samples, HIV antibodies were detected by ELISA in seven out of 22 seropositive individuals. In contrast, testing of 'whole saliva' samples from 79 HIV-seropositive and 115 HIV-seronegative individuals resulted in a 100% correlation with HIV serum status. The positive reaction of 20 'whole saliva' specimens was confirmed in a line immunoassay, whereas in an immunoblot assay only seven specimens were positive, one negative, and 12 indeterminate. In an HIV prevalence study among drug users, 395 'whole saliva' samples were tested in two different ELISA. Both assays showed complete agreement in detecting 58 positive and 337 negative samples. All positive samples were confirmed by the line immunoassay. Conclusion Our study demonstrates that 'whole saliva' specimens are a good alternative to blood samples in epidemiological studies of HIV prevalence in high-risk groups.