To examine trends in predictors of HIV-related mortality among cohorts of persons living with AIDS (PLWA) in New York City (NYC), nine calendar year-specific cohorts of PLWA were created from 1993 to 2001. Cohorts were defined as persons who had been alive at any time during that year and had been diagnosed with AIDS before the end of that year. Predictors of death because of HIV-related causes of death were assessed by examining year-specific, stratified death rates per 1,000 PLWA and adjusted relative risks (RRs) from proportional hazards models. We conducted an analysis of AIDS surveillance data PLWA in NYC between 1993 and 2001. Univariate and multivariate Cox proportional hazards models were constructed for each calendar year cohort to evaluate trends in the RR of HIV-related death over the subsequent 5 years, adjusting for sex, race/ethnicity, age, transmission risk, borough of residence, category of AIDS diagnosis [opportunistic illness (OI) or CD4 count <200 cells/microL], time since AIDS diagnosis, and CD4 count at time of AIDS diagnosis. Death rates due to all causes and HIV-related causes declined substantially during 1993-1997 and then stabilized in all subgroups of PLWA between 1998 and 2001. Beginning in 1995, differences in survival emerged in some subgroups, such that by 2001 (1) injecting drug users (IDUs) had poorer survival compared with men who have sex with men (MSM) [RR(2001) = 2.1, 95% confidence intervals (95% CI) = 1.8-2.4]; (2) black and Hispanic PLWA had a significantly higher risk of death than white PLWA (RR(2001) = 1.4, 95% CI = 1.2-1.6, RR(2001) = 1.2, 95% CI = 1.1-1.4, respectively, and (3) PLWA aged 60 and above had poorer survival compared with younger persons (RR(2001) = 2.4, 95% CI = 1.9-3.0), after adjustment for other factors. The observed disparities that began to emerge in 1995 may be attributable to differential effects of, access to, or usage of highly active antiretroviral therapy (HAART). More targeted studies are needed to determine why such disparities have emerged.