Regional runoff, evapotranspiration, leaf area index (LAI) and potential vegetation were modeled for Mesoamerica using the SVAT model MAPSS. We calibrated and validated the model after building a comprehensive database of regional runoff, climate, soils and LAI. The performance of several gridded precipitation forcings (CRU, FCLIM, WorldClim, TRMM, WindPPT and TCMF) was evaluated and FCLIM produced the most realistic runoff. Annual runoff was successfully predicted (R2=0.84) for a set of 138 catchments with a regression slope of 0.88 and an intercept close to zero. This low runoff bias might originate from MAPSS assumption of potential vegetation cover and to underestimation of the precipitation over cloud forests. The residues were found to be larger in small catchments but to remain homogeneous across elevation, precipitation and land use gradients. Based on the assumption of uniform distribution of parameters around literature values, and using a Monte Carlo-type approach, we estimated an average model uncertainty of 42% of the annual runoff. The MAPSS model was found to be most sensitive to the parameterization of stomatal conductance. Monthly runoff seasonality was fairly mimicked (Kendal tau correlation coefficient higher than 0.5) in 78% of the catchments. Predicted LAI was consistent with EOS-TERRA-MODIS collection 5 and ATSR-VEGETATION-GLOBCARBON remotely sensed global products. The simulated evapotranspiration:runoff ratio increased exponentially for low precipitation areas, stressing the importance of accurately modeling evapotranspiration below 1500 mm of annual rainfall with the help of SVAT models such as MAPSS. We propose the first high resolution (1 km2 pixel) maps combining runoff, evapotranspiration, leaf area index and potential vegetation types for Mesoamerica.