The use of brackish aquifers for aquifer storage and recovery (ASR) projects requires careful examination of the recoverability of potable water since mixing with the native water causes the injected water quality to degrade, resulting in a reduced volume of potable water which can be recovered. The success of ASR injection/extraction wells in brackish aquifers is determined by evaluating the recovery efficiency of the system, the volume of water recovered (before an established threshold concentration is surpassed) divided by the volume of water injected. In conjunction with traditional aquifer tests and borehole geophysics, computer models have become an essential tool for evaluating the logistical feasibility and potential for adequate recovery efficiency of ASR programs. ASR modelers have often characterized aquifers by a single averaged hydraulic conductivity for the entire storage zone in which the water was to be injected. Since real aquifer systems are often heterogeneous in nature, it is reasonable to conclude that computer models generated using the averaged hydraulic conductivity (homogeneous) approach are much less likely to accurately predict the extent injected water will travel and the amount of mixing which will occur, and thus the recovery efficiency of the system. To understand the effects of aquifer heterogeneity and other controlling parameters on ASR recovery efficiency, two models were constructed. One model was designed with a homogeneous permeability distribution and the other with a heterogeneous permeability distribution for the flow zone. The models are based on the San Diego Formation, a vertically and laterally heterogeneous aquifer in southwest San Diego County with a regional hydraulic gradient of 0.002 and typical total dissolved solids (TDS) concentrations ranging from 1,200 milligrams per liter (mg/l) to 3,300 mg/l. A hypothetical ASR project was modeled to simulate cyclic injection and extraction, each lasting for 180 days, for a 5 year period. Numerous simulations were run for both the homogeneous and heterogeneous models to test the effect on recovery efficiency of various aquifer parameters including native water quality, dispersivity, vertical hydraulic conductivity, and permeability contrast, as well as to examine design parameters such as pumping rate and the inclusion of a storage period between injection and extraction cycles. Results of the three-dimensional flow and solute-transport models, obtained using MODFLOW-SURFACT, were used to produce cross-sectional figures of the predicted distributions of injected water and residual injected water following extraction. Examination of these figures showed that the primary source of the extracted water was the up-gradient supply of low concentration water, indicating the strong effect of the regional hydraulic gradient. Also apparent from these figures, is that a significant amount of injected water moved vertically into the leaky confining layers of the aquifer, and that this water was less easily swept down-gradient by the regional hydraulic gradient, suggesting these low permeability layers serve as additional source areas for the extracted low concentration water. Low TDS concentration water injected into the model with a heterogeneous distribution of permeability was concentrated in the lower portion of the screened section of the aquifer where the most permeable layer was located. Increasing the hydraulic gradient of the heterogeneous model resulted in much of the injected water effectively escaping down-gradient beyond the capture zone of the well. Similarly, increasing the hydraulic conductivity of the most permeable layer of the heterogeneous model resulted in loss of injected water down-gradient due to its preferential movement into the high permeability layer. The addition of storage periods resulted in less of the injected water stored up-gradient due in part to the smaller volume of water injected but also due to the time available for the water to move down-gradient of the well before commencement of the extraction period. Increasing the pumping rate resulted in a significantly larger plume of low TDS water both vertically and laterally. Results of the models showed that recovery efficiencies are sensitive to several parameters. Recovery efficiencies are overwhelmingly dependent upon the native water quality of the aquifer. A change in the native water quality from a TDS concentration of 1,200 mg/l to 3,000 mg/l resulted in approximately half the recovery efficiency. Regional hydraulic gradients also significantly control recovery efficiencies of ASR projects. Increasing the hydraulic gradient from 0.002 to 0.005 resulted in approximately one-quarter less recovery efficiency. The addition of vertically heterogeneous layers within the flow zone, with hydraulic conductivities ranging from 1 ft/day to 205 ft/day, resulted in a minor decrease in the recovery efficiency of approximately 8 percent after the final cycle. To further evaluate the possible implications of heterogeneity the heterogeneous model was altered by increasing the hydraulic conductivity of the most permeable layer from 205 ft/day to 500 ft/day. This change resulted in a more significant decrease in the recovery efficiency of 16 percent, relative to a homogeneous model with an equivalent transmissivity, suggesting that the hydraulic conductivity contrast between layers is also a controlling factor on ASR recovery efficiency. The addition of storage periods between the injection and extraction periods only slightly decreased the recovery efficiency. Conversely, doubling the pumping rate only slightly increased the recovery efficiency. Changes in the dispersivity values and vertical hydraulic conductivities had little to no affect. The results presented here do not entirely agree with results presented in two previous studies which similarly examined the effect of heterogeneity on recovery efficiency, but used the solute-transport model MT3D. The models in this study were modified to allow comparison with the results of those studies, and in both cases the recovery efficiencies predicted in the MT3D models were approximately 30 percent below those predicted by MODFLOW-SURFACT. It has previously been documented that early versions of the MT3D code are not mass conservative which may be one explanation for the disparity. The results of this study should encourage ASR project designers to carefully assess native groundwater quality, regional hydraulic gradient and vertical heterogeneity of a prospective brackish aquifer through fieldwork before attempting to model an ASR project upon which the project’s feasibility will be judged.