The accuracy of a saliva collection and testing protocol for determination of HIV-1 antibody status was assessed under realistic field conditions. The 1,256 study participants came from two lower-prevalence settings--a self-referral testing clinic (478) and a street-based outreach program (431)--and two high-prevalence clinics--an HIV/AIDS treatment clinic (337) and a hemophilia clinic (10). Saliva was collected using the Omni-Sal collection device and was tested by technicians blind to serum status using a modified protocol that employed the Recombigen HIV-1 EIA. Serum was tested using standard methods. A single saliva enzyme immunoassay (EIA) correctly identified 358 of 368 seropositive individuals, for a sensitivity of 97.3% (95% CI: 95.1%-98.7), and 888 of 888 seronegative individuals, for a specificity of 100% (95% CI: 99.6-100). Confirmation of saliva EIA positives with a repeat saliva EIA and an in-house saliva radioimmunoprecipitation assay resulted in a substantial drop in sensitivity to 85.3% (95% CI: 81.0-89.0). We conclude that a single saliva EIA using the modified test protocol described is sufficiently accurate for surveillance purposes, but we do not recommend it for diagnostic purposes or screening.