Since the introduction of zidovudine, a nucleoside analogue reverse transcriptase inhibitor, there has been continuous improvement in the efficacy of antiretroviral treatments. Briefly, antiretroviral therapy moved from the era of zidovudine monotherapy to a combination of two analogues of the reverse transcriptase and then to triple-drug therapy with the introduction of protease inhibitors in 1996. The efficacy of triple-drug combinations was further consolidated with the introduction of the non-nucleoside analogue reverse transcriptase inhibitor. Beyond 1996, the impressive improvement in efficacy has implied that death and clinical endpoints could not be reasonably used as primary efficacy endpoints. Recent studies used plasma viral load as surrogate endpoints to evaluate efficacy of antiretroviral combinations. Progress in antiretroviral efficacy has also led to a move from superiority to noninferiority design. Methodological issues in noninferiority design have been already criticized, but recent trials provide a good opportunity for discussing some key features of that design. Recent studies are used to illustrate some inconsistencies in the choice of response rates, power and noninferiority margin. It appears that HIV noninferiority trials are overpowered by assuming lower success rates than those observed, enrolling a large number of patients and choosing a large margin. Consequently, failure to demonstrate noninferiority is uncommon. Novel designs or endpoints should be introduced emphasizing the expected benefits in terms of toxicity, adherence, resistance or costs.