Study methods of assessing HIV trends in sexually transmitted disease (STD) clinics where blinded seroprevalence has been among the highest in the United States. Cross-sectional and retrospective cohort. Reviewed computerized records of the four Miami STD clinics for 1987-1992. A total of 53,403 persons had 70,793 tests. When testing began, 13% were HIV-positive. By 1992, 35% of all tests were performed on previously HIV-negative persons. This caused a faster decline in overall seroprevalence (to 7%) compared with seroprevalence among persons tested for the first time (to 9%). The percentage of tests performed for previously HIV-negative persons varied among age, race and sex subgroups, causing confounding if overall seroprevalence was used to compare groups. Seroconversion occurred in 514 patients. Black women were most likely to be positive on retesting (4.3%). Incidence rates did not change much over time, remaining at 2.5% per year. When using a routine HIV-testing data to analyze trends, separate analyses should be undertaken for patients who were (or were not) previously tested. In these patients, the prevalence decreased slightly but incidence remained high, particularly among black women.