Model-based studies of antiviral use to mitigate the impact of moderate and severe influenza pandemics implicitly take the viewpoint of a central public health authority. However, it seems likely that the key decision of when to use antivirals will be made at the household level. We used a stochastic compartmental model of the transmission of influenza within and between households to evaluate the expected mortality under two strategies: households saving available antivirals for treatment only and households implementing prophylaxis as well as treatment. Given that every individual in the population was allocated a single course of antivirals, we investigated the impact of these two strategies for a wide range of AVED, the efficacy of antivirals in preventing death in severe cases (AVED=1 for complete protection). We found a cross-over point for our baseline parameter values in a regime where antivirals were still highly effective in reducing the chance of death: below AVED=0.9 the optimal strategy was for households to use both treatment and prophylaxis. We also considered the possibility that a small number of households might "cheat" by choosing to follow the treatment-only strategy when other households were following treatment with prophylaxis. The cross-over point for cheating households was considerably lower, at AVED=0.6, but substantially above 0. These results suggest that unless antivirals are almost completely effective in reducing the chance of death in serious cases, households will likely be better served implementing prophylaxis as well as treatment. More generally, our study illustrates the potential value of considering viewpoints other than a central authority when conducting model-based analysis of interventions against infectious disease.