Lake Starr and other lakes in the mantled karst terrain of Florida's Central Lake District are surrounded by a conductive surficial aquifer system that receives highly variable recharge from rainfall. In addition, downward leakage from these lakes varies as heads in the underlying Upper Floridan aquifer change seasonally and with pumpage. A saturated three-dimensional finite-difference ground-water flow model was used to simulate the effects of recharge, Upper Floridan aquifer heads, and model time scale on ground-water exchange with Lake Starr. The lake was simulated as an active part of the model using high hydraulic conductivity cells. Simulated ground-water flow was compared to net ground-water flow estimated from a rigorously derived water budget for the 2-year period August 1996-July 1998. Calibrating saturated ground-water flow models with monthly stress periods to a monthly lake water budget will result in underpredicting gross inflow to, and leakage from, ridge lakes in Florida. Underprediction of ground-water inflow occurs because recharge stresses and ground-water flow responses during rainy periods are averaged over too long a time period using monthly stress periods. When inflow is underestimated during calibration, leakage also is underestimated because inflow and leakage are correlated if lake stage is maintained over the long term. Underpredicted leakage reduces the implied effect of ground-water withdrawals from the Upper Floridan aquifer on the lake. Calibrating the weekly simulation required accounting for transient responses in the water table near the lake that generated the greater range of net ground-water flow values seen in the weekly water budget. Calibrating to the weekly lake water budget also required increasing the value of annual recharge in the nearshore region well above the initial estimate of 35 percent of the rainfall, and increasing the hydraulic conductivity of the deposits around and beneath the lake. To simulate the total ground-water inflow to lakes, saturated-flow models of lake basins need to account for the potential effects of rapid and efficient recharge in the surficial aquifer system closest to the lake. In this part of the basin, the ability to accurately estimate recharge is crucial because the water table is shallowest and the response time between rainfall and recharge is shortest. Use of the one-dimensional LEACHM model to simulate the effects of the unsaturated zone on the timing and magnitude of recharge in the nearshore improved the simulation of peak values of ground-water inflow to Lake Starr. Results of weekly simulations suggest that weekly recharge can approach the majority of weekly rainfall on the nearshore part of the lake basin. However, even though a weekly simulation with higher recharge in the nearshore was able to reproduce the extremes of ground-water exchange with the lake more accurately, it was not consistently better at predicting net ground-water flow within the water budget error than a simulation with lower recharge. The more subtle effects of rainfall and recharge on ground-water inflow to the lake were more difficult to simulate. The use of variably saturated flow modeling, with time scales that are shorter than weekly and finer spatial discretization, is probably necessary to understand these processes. The basin-wide model of Lake Starr had difficulty simulating the full spectrum of ground-water inflows observed in the water budget because of insufficient information about recharge to ground water, and because of practical limits on spatial and temporal discretization in a model at this scale. In contrast, the saturated flow model appeared to successfully simulate the effects of heads in the Upper Floridan aquifer on water levels and ground-water exchange with the lake at both weekly and monthly stress periods. Most of the variability in lake leakage can be explained by the average vertical head difference between the lake and a re