Airflow models of buildings require dozens to hundreds of parameter values, depending on the complexity of the building and the level of fidelity desired for the model. Values for many of the parameters are usually subject to very large uncertainties (possibly an order of magnitude). Experiments can be used to calibrate or "tune" the model: input parameters can be adjusted until predicted quantities match observations. However, experimental time and equipment are always limited and some parameters are hard to measure, so it is generally impractical to perform an exhaustive set of measurements. Consequently, large uncertainties in some parameters typically remain even after tuning the model. We propose a method to help determine which measurements will maximally reduce the uncertainties in those input parameters that have the greatest influence on behavior of interest to researchers. Implications for experimental design are discussed.