The paper deals with the Modifiable Areal Unit Problem in residential location choice. The household location choice model from the UrbanSim simulation framework is calibrated at different spatial scales. The capacity to predict the geographical distribution of population is a criterion for the choice of an areal unit. Thus, residential location choice is predicted for past years with the same model specification at the scales of blocks, zones, and municipalities. The municipality-based model has better predictive capacity and the least stochastic variation in comparison with the block-based and zone-based models, but the block-based output aggregated to municipal scale produces the best prediction comparable with an evenly split population growth.