Each region has developed its own land-use system and factors prescribing land use are more or less different in every region. Previous research suggests that an ordinary regression model is an inadequate tool to identify the factors if the study area includes various regions. To elucidate the factors influencing farmland distribution in Japan, the author applied a multilevel model. It has been invented for analyzing hierarchically structured data, which consist of lower-level samples nested within higher level(s). It enables us to estimate not only the influences of each factor at lower levels but also how much they vary at higher levels. (C) 2001 Elsevier Science Ltd. All rights reserved.