Reasonably airtight buildings can protect occupants from large-scale outdoor airborne releases. However, some houses are leaky, as air tightness tends to vary greatly in a housing stock. We modeled the health consequences if a single-family residential community were to "shelter-in-place," for two different models of a toxic release: (I) a simple Gaussian puff, and (II) a realistic simulation of outdoor transport and dispersion generated by the National Atmospheric Release Advisory Center. We predicted the health effects under two different assumptions: (1) every house has the same indoor-outdoor air-exchange rate, or (2) the houses have a lognormal distribution of air-exchange rates. The assumption that every house has the same air-exchange rate (at the median of the actual distribution) can lead to an under-prediction of the community area adversely affected by the release by a factor of 3 or more. The difference is largest if the dose-response relationship of the chemical is highly nonlinear.