We present a novel image-based technique to estimate a subject-specific boundary condition (BC) for computational fluid dynamics (CFD) simulation of pulmonary air flow. The information of regional ventilation for an individual is derived by registering two computed tomography (CT) lung datasets and then passed to the CT-resolved airways as the flow BC. The CFD simulations show that the proposed method predicts lobar volume changes consistent with direct image-measured metrics, whereas the other two traditional BCs (uniform velocity or uniform pressure) yield lobar volume changes and regional pressure differences inconsistent with observed physiology.