Human immunodeficiency virus (HIV)-infected subjects receiving zidovudine were randomized either to add stavudine (d4T) or didanosine (ddI) to their current regimen or to switch to ddI or d4T monotherapy. After 16 weeks of therapy, the mean reduction in HIV RNA from baseline was 0.14 log(10) copies/mL in patients receiving d4T or zidovudine plus d4T. In subjects receiving ddI or ddI plus zidovudine, reductions were 0.39 and 0.56 log(10), respectively. CD4 cell counts remained stable or showed modest increases in all arms except the zidovudine plus d4T arm. Patients receiving zidovudine plus d4T showed progressive declines in CD4 cell counts with a median of 22 cells/mm(3) below baseline by 16 weeks. Examination of intracellular levels of d4T-triphosphate in 6 subjects was consistent with previous in vitro studies demonstrating pharmacologic antagonism between zidovudine and d4T. Analysis of these data suggests that zidovudine and d4T should not be prescribed in combination and that ddI provides greater antiviral activity than d4T in zidovudine-treated patients.