Background Patients with advanced HIV infection at the time of diagnosis and patients not responding to antiretroviral therapy are at risk of cytomegalovirus (CMV) disease. Earlier studies of patients with HIV infection have demonstrated that the diagnosis is often first made post-mortem. In recent years new molecular biological tests have become available for diagnosis of CMV disease. Although clinical evaluation of tests for diagnosis of CMV disease in HIV-infected individuals is suboptimal without autopsy, no results from such studies have been published. The aim of this study was to explore the diagnostic utility of CMV quantitative polymerase chain reaction (PCR) in plasma from HIV and CMV seropositive patients who died during the period 1991–2002 and in whom autopsy was performed. Methods Autopsy was performed in all cases, as part of routine evaluation of HIV-infected cases followed at Ullevaal University Hospital. Of 125 patients included, 53 had CMV disease, 37 of whom were first diagnosed at autopsy. CMV disease was diagnosed either by ophthalmoscopic findings typical of CMV retinitis, biopsy or autopsy. One or two plasma samples taken prior to the first diagnosis of CMV disease (alive or at autopsy) or death without CMV disease were analysed by CMV quantitative PCR. Sensitivity, specificity, positive and negative predictive values were calculated for different CMV viral load cut-offs and according to detection of viraemia in one versus two samples. Results Twenty-seven of 53 patients with CMV disease (51%) and 10 of 72 patients without CMV disease (14%) had detectable viraemia in at least one sample. Sensitivity and negative predictive value (NPV) of the test, maximised with a cut-off at the test's limit of detection of CMV viraemia (400 copies/mL), were 47% and 70%, respectively. With cut-off at 10 000 copies/mL, specificity and positive predictive value (PPV) were 100%. With a requirement for CMV viraemia in two samples, specificity and PPV were 100% in patients with CMV viraemia above the limit of detection. Conclusion Our results indicate that quantitative CMV PCR is best used to rule in, rather than to rule out CMV disease in HIV-infected individuals at high risk.