Distribution of human suffering due to earthquake depends largely on the spatial distribution of population at the moment the earthquake occurs. For this, a computational model for predicting day-long spatio-temporal distribution of commuters was developed by using a number of nation-wide statistical data such as “Population Census of Japan”, “Survey on Time Use and Leisure Activities”, and “Establishment and Enterprise Census”. The model predicts behavior of individual commuters by considering their attributes including gender, occupation, place of work and residence, and whether he/she works at home or not. As a case study, the proposed model was applied to the Keihanshin Metropolitan Area for estimating distribution of commuters, and those who unable to return home in case of the wide-area traffic interruption due to earthquake.