Most studies of landscape effects on plants in forests have explored only relatively small landscape scales (size of spatial extent) considering the poor-dispersal ability generally reported for forest plants, or failed to verify confidently that the scales at which landscape is considered best predicts plant patterns. We. investigated the effect of landscape composition on response of plant communities according to three scales - 400 ha, 100 ha and 25 ha - in 263 forests in the northern half of France. We hypothesized that landscape composition at the 400-ha scale better predicted plant responses than smaller scales. We built models to compare the effects of each scale on individual species response and plant composition of 1902 700-m(2) vegetation plots, considering the shared and pure effects of scales. Only landscape composition at the 400-ha scale showed a critical pure effect on plant composition. Similarly, the 400-ha scale better predicted individual species response. Using a large dataset, we brought evidence that landscape composition further than 100 ha and at least up to 400 ha had more effects on local plant diversity in forests, which means that smaller-scale investigations may miss the influence of landscape on plant patterns. The effects of large spatial scale may reflect the legacies of past landscape or the long-distance dispersal capacity of plants. Our findings call for better consideration of the large-scale effects of landscape on plants both in scientific studies and management plans.