We analyzed trends over time and determinants of late diagnosis of HIV infection among people diagnosed with AIDS in 1986 to 1998 in a tertiary care center in Rome, Italy. Information on the date of a first HIV test was collected prospectively, in addition to data routinely collected for AIDS reporting. Patients with AIDS were defined as "late testers" if the time interval between first positive HIV test result and AIDS diagnosis was < or = 3 months. Overall, 503 people with AIDS of 1977 included in the analysis (25.4%) were late testers. the proportion of late testers decreased from 62.5% in 1986 to 16% in 1995. Thereafter, this proportion increased to 20.5% in 1996, 33.7% in 1997, and 36.6% in 1998. In multivariate analysis, the following variables were significantly associated with late testing: AIDS diagnosis in years 1986 to 1993 or 1997 to 1998 compared with 1995, male gender, age > or = 45 years, men who have sex with men, heterosexual contacts, or having unknown transmission mode compared with intravenous drug users, and being born outside Italy. Since 1996, the overall number of AIDS cases diagnosed at our center began to decrease whereas the number of late-testing AIDS patients did not decrease, resulting in an increasing proportion of late testers during the last 3 years of the study. This findings may reflect the effect of combination antiretroviral therapy in slowing progression to AIDS of HIV-infected persons aware of their status. A relevant number of people still discover their HIV infection late and may therefore miss treatment opportunities. New testing strategies are needed to reach more people who engage in high-risk behaviors, especially those at risk for sexual transmission, and those born outside Italy.