We analyzed more than 70,000 air leakage measurements in houses across the United States to relate leakage area—the effective size of all penetrations of the building shell—to readily available building characteristics such as building size, year built, geographic region, and various construction characteristics. After adjusting for the lack of statistical representativeness of the data, we found that the distribution of leakage area normalized by floor area is approximately lognormal. Based on a classification tree analysis, year built and floor area are the two most significant predictors of leakage area: older and smaller houses tend to have higher normalized leakage areas than newer and larger ones. Multivariate regressions of normalized leakage are presented with respect to these two factors for three house classifications: low-income households, energy program houses, and conventional houses. We demonstrate a method of applying the regression model to housing characteristics from the American Housing Survey to derive a leakage-area distribution for all single-family houses in the US. The air exchange rates implied by these estimates agree reasonably well with published measurements.