25 patients with AIDS or serious HIV infection were asked to specify the minimum time zidovudine should prolong life before they would be willing to take the drug for the rest of their lives. Most patients were unable to state a least acceptable survival benefit. Hence, the clinical relevance of sample size calculations in AIDS trials may be questioned since a survival benefit of zero would lead to an infinite sample size. Alternatively, one could aim first at a fixed number of deaths. Thereafter, it could be calculated how many more years the study should continue to secure that a mortality difference of some universally adopted magnitude would not be overlooked. This approach would not solve the sample size dilemma but would convert it into a pragmatic question: are we willing to continue the study for another x number of years?