To estimate the prevalence of HIV by anonymous saliva testing in Her Majesty's Prison, Saughton (Saughton Prison), Edinburgh, UK. To elicit linked anonymous risk factor information from which to estimate risk scores for those who had taken an HIV blood test and, among drug injectors, for those who were HIV-1-antibody-positive on saliva testing. Saughton Prison on 15 and 16 August 1991; HIV Immunology and Regional Virus Laboratories, Edinburgh, and the Medical Research Council Biostatistics Unit, Cambridge, UK. Male inmates (378 out of a total of 499) of Saughton Prison. Answers to a brief questionnaire about age, usual residence, present and past custodial sentences, drug injecting and sexual behaviour prior to and in prison, HIV testing and history of acute hepatitis. HIV-1-antibody status was established by saliva testing. Eighteen per cent of participants were injecting drug users (IDU), of whom approximately one-half (47%) had injected while inside prison. Ninety men (26%), including 40 (14%) of 278 participants who had never injected drugs and 77% of IDU participants, had taken an HIV blood test. Nine per cent of all participants and 35% of IDU participants had had an acute attack of hepatitis. Forty-one (62%) of 66 IDU had been imprisoned five or more times before their current prison sentence. After taking account of region of residence, injecting drug history and acute hepatitis, aspects of sentencing and sexual behaviour were not determinants of those who had been tested for HIV. On the study days, 18 out of 499 (3.6%) participants were known to prison medical officers to be HIV-infected. Following saliva testing, HIV prevalence was 17 out of 375 (4.5%) inmates tested. All 17 had at some time 'taken the blood test for HIV' and all had injected non-medically prescribed drugs. Edinburgh residence, age 26-30 years, have injected in prison and having first injected before 1983 all contributed to the risk score for whether an IDU was HIV-1-antibody-positive on saliva testing. Documented HIV prevalence in saliva was 4.5%, which--assuming no volunteer bias (as supported by questionnaire returns)--suggests that actual HIV prevalence was 25% greater than revealed to Saughton's prison medical service. All 17 inmates who were HIV-1-antibody-positive on saliva testing had injected non-medically prescribed drugs. The high reported frequency by inmates of injecting in prison highlights the urgent requirement for drug reduction and rehabilitation programmes for injecting inmates. Linked anonymous voluntary HIV testing of saliva can provide valuable information about HIV prevalence for the planning of prison resources and policy.