A historical cohort study was conducted in three public AIDS referral services in Belo Horizonte, Minas Gerais, to assess the relationship between health care utilization and patient survival following AIDS diagnosis. A review of medical and laboratory records was performed for HIV-infected patients seeking care for the first time in 1989-92. Among 291 patients initially categorized as 'A' [asymptomatic, acute HIV, or persistent generalized lymphadenopathy-PGL] or 'B' [symptomatic, non-'A', or AIDS-indicator conditions] (CDC, 1992) and who progressed to AIDS, 57.0% died. Mortality rate was 34.9 person-months. Overall median survival time following AIDS diagnosis was 14.3 months. Multivariate analysis showed that lack of AZT use (RR=1.87; 95% CI=1.34-2.61), advanced initial staging (RR=1.68; 95% CI=1.20-2.35), 9 or more inpatient days (RR=1.55; 95% CI=1.11-2.17), and intervals between outpatient visits longer than 6 months (RR=0.30; 95%CI= 0.16-0.56) were associated with death. The analysis suggests that: Patients who used health services more often had poorer prognosis; Patients who received AZT survived longer than those who did not; and variables used to assess health care utilization actually express the end of a process involving seeking and obtaining care.