Many tropical islands have limited water resources with historically increasing demand, all potentially affected by a changing climate. The effects of climate change on island hydrology are difficult to model due to steep local precipitation gradients and sparse data. This work uses 10 statistically downscaled general circulation models (GCMs) under two greenhouse gas emission scenarios to evaluate the uncertainty propagated from GCMs in projecting the effects of climate change on water resources in a tropical island system. The assessment is conducted using a previously configured hydrologic model, the Precipitation Runoff Modelling System (PRMS) for Puerto Rico. Projected climate data and their modelled hydrologic variables versus historical measurements and their modelled hydrologic variables are found to have empirical distribution functions that are statistically different with less than 1 year of daily data aggregation. Thus, only annual averages of the projected hydrologic variables are employed as completely bias‐corrected model outputs. The magnitude of the projected total flow decreases in the four regions covering Puerto Rico, but with a large range of uncertainty depending on the makeup of the GCM ensemble. The multi‐model mean projected total flow decreases by 49–88% of historical amounts from the 1960s to the 2090s for the high emissions scenarios and by 39–79% for the low emissions scenarios. Subsurface flow contributions decreased the least and groundwater flow contributions decreased the most across the island. At locations critical to water supply for human use, projected streamflow is shown to decrease substantially below projected withdrawals by 2099.