We studied predictors for losses to follow-up and the impact of such losses in the AIDS Clinical Trials Group 019 protocol, a long-term randomized trial of immediate versus deferred antiretroviral therapy in asymptomatic HIV-1-infected patients with >500 CD4 cells/mm3. The trial was selected because of its key importance in determining guidelines for antiretroviral therapy, and because it had the longest follow-up among all antiretroviral trials and the largest percentage of patients whose vital status was unknown at study end. Younger age, a history of parenteral drug use, and nonwhite race were associated with higher rates of loss to follow-up, but race was not an important predictor after adjusting for clinical site. There was large and statistically significant variability in the rates of losses among different clinical sites (p < 0.001). Patient retention was significantly better in clinical sites that enrolled many participants, with 25% of enrollees lost to follow-up in sites enrolling >100 patients and 44% in sites enrolling <33 patients each. As a group, patients lost to follow-up after the 2nd year had steeper declines of CD4 cell counts, and a significantly larger proportion had reached a CD4 cell count <300/mm3 in the year before being lost, compared with patients remaining in the study. Losses to follow-up probably decreased substantially the observed number of primary endpoints, curtailed the power of the trial to demonstrate any difference between immediate and deferred initiation of antiretroviral therapy, and may have introduced large bias in the estimated hazard ratio for the primary endpoint and its statistical significance.