While the neuraminidase (NA) inhibitor oseltamivir is currently our first line of defense against a pandemic threat, there is little information about whether in vitro testing can predict the in vivo effectiveness of antiviral treatment. Using a panel of five H5N1 influenza viruses (H5 clades 1 and 2), we determined that four viruses were susceptible to the drug in vitro (mean 50% inhibitory concentration [IC(50)], 0.1 to 4.9 nM), and A/Turkey/65-1242/06 virus was slightly less susceptible (mean IC(50), 10.8 nM). Two avian viruses showed significantly greater NA enzymatic activity (V(max)) than the human viruses, and the five viruses varied in their affinity for the NA substrate MUNANA (K(m), 64 to 300 muM) and for oseltamivir carboxylate (K(i), 0.1 to 7.9 nM). The protection of mice provided by a standard oseltamivir regimen (20 mg/kg/day for 5 days) also varied among the viruses used. We observed (i) complete protection against the less virulent A/chicken/Jogjakarta/BBVET/IX/04 virus; (ii) moderate protection (60 to 80% survival) against three viruses, two of which are neurotropic; and (iii) no protection against A/Turkey/65-1242/06 virus, which induced high pulmonary expression of proinflammatory mediators (interleukin-1alpha [IL-1alpha], IL-6, alpha interferon, and monocyte chemotactic protein 1) and contained a minor subpopulation of drug-resistant clones (I117V and E119A NA mutations). We found no correlation between in vitro susceptibility and in vivo protection (Spearman rank correlation coefficient rho = -0.1; P > 0.05). Therefore, the in vivo efficacy of oseltamivir against highly pathogenic H5N1 influenza viruses cannot be reliably predicted by susceptibility testing, and more prognostic ways to evaluate anti-influenza compounds must be developed. Multiple viral and host factors modulate the effectiveness of NA inhibitor regimens against such viruses and new, more consistently effective treatment options, including combination therapies, are needed.