How fuel influences fire spread at different spatial scales has been broadly studied but it is still under research. Although prior research has generally explored fuel effects at the stand scale, there is increasing recognition that forest structure and composition at the landscape scale can regulate fire size when weather conditions allow fire to have enough initial capacity to spread. Simulation models have usually been adopted to test forest landscape structure’s effect on fire activity, but we still lack empirical evidence of these effects on a broad scale and using multiple fires. In this study, we use the graph theory and the percolation theory to test the relationship between forest functional connectivity and fire size in a Mediterranean region (Catalonia, NE Spain, ~32000 km2) and under different weather conditions. Using a total sample of 554 fires that occurred between 1987 and 2019, we relate fire size with forest connectivity inside the fire perimeter, which we describe with different interpretations of forest-land cover maps and at different fire dispersal distances (10-, 100- and 2000- meters spotting capacity). Previously, fires were classified according to the weather and drought conditions in which they occurred (windy, arid or regular days, and dry or non-dry conditions) with the hypothesis that the capacity of fire to “read” the forest landscape differs according to weather variables. We found that a forest percolation threshold existed for most types of fires: forest connectivity had a positive influence on fire size until a threshold at which the landscape becomes continuous and other processes to control fire cease to have effect. This percolation threshold is about 0.40 for fires occurring under arid conditions. Under regular weather conditions, the threshold is non-existent and forest connectivity always exerts a positive effect on fire size. Notably, forest connectivity has no influence on fire size on windy- and dry-weather days. All these results are of paramount relevance for forest and landscape management aimed at reducing wildfire risk, since for fires driven by extreme weather conditions (more prone to happen in the future), landscape offers less opportunities to stop them. Researchers and managers need to identify the probability of experiencing each of the different weather types in a climate change context to boost the management of forests and landscapes in a cost-effective way.