Numerical modeling is an important tool for assessing water resources and evaluating contaminant transport in aquifers. The critical issue for many karst systems is maintaining spring flow during periods of drought; therefore, simulation of low flow periods is critical for water resources management. The purpose of this study was to evaluate the ability of distributed and lumped models to simulate water flow in a karst aquifer. The models were developed for the Barton Springs segment of the Edwards Aquifer, Texas. The MODFLOW code was used for the distributed model, which consisted of a uniform grid (150 m x 305 m) aligned parallel to the dominant fault structure. The lumped model consisted of 6 cells representing each of the river basins in the aquifer. Steady state simulations were used to determine the spatial variability in hydraulic conductivity. Transient simulations were conducted for 1989 through 1998 that included periods of low and high flow. Both models fairly accurately simulated the temporal variability in spring discharge; therefore, if the focus of the model is spring discharge, either distributed or lumped approaches can be used. The distributed model provides much more detail on spatial variability in hydraulic head and generally reproduced the hydrographs in monitoring wells with the exception of those that penetrated conduits and had constant water levels. While the general impact of the amount of pumping on spring discharge can be evaluated with a lumped model, more detailed evaluation of pumping requires a distributed modeling approach. Both distributed and lumped models can be used to evaluate the impact of nonpoint source contamination on water quality in the spring but neither model can be used to assess the impact of point source spills and rapid contaminant transport in conduits because the equivalent porous media assumption cannot be used to evaluate conduit flow.