Spatially explicit land-use change prediction is often based on environmental characteristics of land-use types, such as soil type and slope, as observed at one time instant. This approach presumes that relationships between land use and environment are constant over time. We argue that such temporal stationarity is uncommon in many land-use systems, and we demonstrate by means of a case study how this may affect the performance of land-use change inference. For this purpose, the spatial distribution of cropland and uncultivated land in Mediterranean Europe in 2000 was studied, as well as recent land-use changes from cropland to uncultivated land and vice versa between 1990 and 2000. Both land-use and land-use change frequencies were captured in regression equations that related them to environmental characteristics. A comparison between these two regression types suggests that many traditional constraints (slope, soil quality, accessibility) have become less important, while climatic variables have gained importance. These results can be explained by lower costs of constraints due to technological and economic developments, and by changes in relative average revenues of different land uses.