Grasslands in Japan are decreasing in area and becoming increasingly fragmented. To understand how this will affect grassland organisms, it is important to predict species distributions on a landscape scale. Large-scale models are well suited to identifying potential habitats on a regional scale, but the applicability of such models across different landscapes is not clear. In previous studies of the harvest mouse (Micromys minutus), local-scale species distribution models (SDMs) have been devised for some areas; however, no studies have compared the applicability of such models between different landscapes. Here, we construct an extensive SDM fitted to two districts of western Japan as well as local models for each district. We also verified the accuracy of these extensive models and the transferability to local models. Extensive models using different landscapes for harvest mice yielded acceptable predictions. Rice paddies and forest areas within a 500m radius and perennial grassland habitats were favored as the predictors of mouse presence in the extensive model. Additionally, it became clear that local models had no transferability to other ranges. In the district with significant amounts of grassland (e.g., rural area), mice preferred larger patches sizes and perennial plant types; however, in the district with fewer grasslands (e.g., urban area), mice initially preferred a larger amount of grassland within a 500-m radius, regardless of patch size. Thus, habitat selection by harvest mice differed between rural and urban areas. These results demonstrate that not only patch quality but also the landscape around the habitat patch are important for habitat selection by the harvest mouse, especially in urban areas with limited grasslands (C) 2016 The Authors. Published by Elsevier B.V.