Adoptive immunotherapy of virus infection with viral-specific CTL has shown promise in animal models and human virus infections and is being evaluated as a therapy for established HIV-1 infection. Defining the individual obstacles for success is difficult in human trials. We have therefore examined the localization, persistence, and antiviral activity of HIV-1 gag-specific CTL clones in both HIV-1-infected and uninfected haplotype-matched human (hu)-PBL-SCID mice. Injection of gag-specific clones but not control CTL into HIV-1-infected hosts reduced plasma viremia by >10-fold but failed to eliminate the virus infection from most treated animals. The failure to eradicate virus did not reflect selection of escape variants because the gag epitope remained unmutated in virus isolates obtained after CTL therapy. Injection of carboxyfluorescein diacetate succinimide ester-labeled CTL demonstrated markedly different fates for gag-specific CTL in the presence or absence of HIV-1 infection. HIV-1-specific CTL rapidly disappeared in infected recipients, whereas they were maintained at high numbers in uninfected mice. By contrast, control CTL were long lived in both infected and uninfected recipients. Thus, interaction of CTL with virus-infected target cells in vivo leads not only to target destruction but also to the rapid disappearance of the infused CTL, and it limits the capacity of CTL therapy to eliminate HIV-1 infection.