The potential impact of pandemic influenza makes effective measures to limit the spread and morbidity of virus infection a public health priority. Antiviral drugs are seen as essential requirements for control of initial influenza outbreaks caused by a new virus, and in pre-pandemic plans there is a heavy reliance on drug stockpiles. The principal target for these drugs is a virus surface glycoprotein, neuraminidase, which facilitates the release of nascent virus and thus the spread of infection. Oseltamivir (Tamiflu) and zanamivir (Relenza) are two currently used neuraminidase inhibitors that were developed using knowledge of the enzyme structure. It has been proposed that the closer such inhibitors resemble the natural substrate, the less likely they are to select drug-resistant mutant viruses that retain viability. However, there have been reports of drug-resistant mutant selection in vitro and from infected humans. We report here the enzymatic properties and crystal structures of neuraminidase mutants from H5N1-infected patients that explain the molecular basis of resistance. Our results show that these mutants are resistant to oseltamivir but still strongly inhibited by zanamivir owing to an altered hydrophobic pocket in the active site of the enzyme required for oseltamivir binding. Together with recent reports of the viability and pathogenesis of H5N1 (ref. 7) and H1N1 (ref. 8) viruses with neuraminidases carrying these mutations, our results indicate that it would be prudent for pandemic stockpiles of oseltamivir to be augmented by additional antiviral drugs, including zanamivir.