The prediction of indoor airflow is indispensable in evaluating the thermal sensation of occupants in a cross-ventilated space because enhanced heat loss due to the elevated convective heat transfer in the occupied zone is an essential part of improving thermal comfort. A domain decomposition technique was developed to separate CFD for internal cross-ventilation flow from external flow outside buildings, and to predict indoor airflow with reasonable computational effort and sufficient accuracy. Decoupling of the flow fields inside and outside the building not only improves efficiency of CFD but also enables optimization of opening positions without taking the external flow field into account. Indoor airflow fields of a cross-ventilated building calculated with CFD using the domain decomposition technique were in good agreement with CFD results for full flow field simulation and measurements by wind tunnel experiment. Case studies were conducted to evaluate cross-ventilation performance of a variety of opening positions under the meteorological conditions of Tokyo and Osaka, and it was found that the relative air velocity in the occupied zone to the air velocity passing through the opening was mainly due to the opening positions.