In the framework of land use changes in the Mediterranean area, I asked to what extent different landscape structures might determine long-term dynamics in Mediterranean ecosystems. To answer this question, a spatially explicit model was developed (the Melca model), incorporating two functional types of woody species dominant in Mediterranean ecosystems: a resprouter (R) and a non-resprouter fire-recruiter (seeders, S). The model was used as a tool for generating hypotheses on the possible consequences of different landscape scenarios. Thus, five different hierarchically structured random landscapes were generated, all having the same cover for the two functional types but different landscape structure (ranging from highly heterogeneous to homogeneous landscapes). After a 100-yr simulation, plant cover and spatial pattern had changed and the changes were different for the different initial spatial configurations, suggesting that long-term vegetation dynamics is spatially dependent (the resultant dynamics are sensitive to the initial spatial structure). In the landscapes where R-type species had a low number of large patches and S-species had a large number of small patches, the number of R-patches increased and their size decreased, while the number of S-patches decreased. In these cases, the final cover of the two types changed little from the initial cover. Landscapes with a large number of small R-patches interspersed with S-patches had a decrease in the number of R-patches, an increase in the number of S-patches and a decrease in the size of S-patches. In these landscapes, final cover was significantly changed, increasing in R-type and decreasing in S-type species. These results suggest that low spatial autocorrelation (low aggregation) favours R-type species. Implications for land management are also discussed.