Fire is a key driver of the structure of plant communities. Fuel traits are important determinants of fire behavior in tropical savannas and, thus, of the effects of fire. However, whether these traits are predictable and how they are influenced by biotic and abiotic drivers remain to be rigorously evaluated. We tested the hypothesis that fuel traits are mutually correlated and change in response to abiotic factors and fire history. We sampled 31 plots in Serra da Canastra National Park (Brazil) distributed in five soil classes and measured the following fuel traits: fuel loads, height, continuity, bulk density, and bed flammability. We determined soil clay content, fire history, climate, canopy cover and elevation. We also obtained data on future fire frequency. To test for correlations among fuel traits, we used Pearson correlation tests and a principal component analysis. For the effect of abiotic factors on the traits and principal component axes, and for the effect of fuel traits on future fire regime, we used generalized linear models. We found two leading axis of fuel trait variability. The first was positively correlated to fuel height, continuity, loads, bed flammability, and grass cover, and was predicted positively by time since last fire. The second axis was positively correlated to fuel bulk density and continuity, shrub and litter cover, negatively to fuel bed flammability, and was predicted by canopy cover and soil clay content (positive associations). Grass fuel loads were the best predictor of future fire frequency and were negatively correlated to canopy cover and positively to time since last fire. Our results suggest that fuel traits change predictably in space and time, that canopy cover and time since last fire are the main factors controlling community flammability, and that grass fuel cover is a key predictor of fire frequency in savannas.