Recirculating well systems provide an engine for the in situ treatment of subsurface contaminants. Although numerous recirculating wells have been installed in the field, for such systems, there is a paucity of comprehensive monitoring data and models constrained to data appearing in the research literature. Here we present an extensive data set combined with detailed inverse and simulation analyses for a two-well groundwater recirculation system used for in situ bioremediation at Edwards Air Force Base in southern California. The conveyor belt flow system, which was established for in situ treatment of trichloroethylene (TCE) in two bioactive zones, was created by pumping water upward in one well and downward in another well, each well being screened in both the upper and lower aquifers. A bromide tracer test was conducted and extensively monitored for 60 days. Combined inverse analysis was conducted on hydraulic heads from 38 monitoring wells, 32 bromide concentration histories, and a constraint on the degree of recirculation that was based on TCE concentration data. Four different formulations involving alternative weighting schemes used in a nonlinear weighted least squares simulation-regression analysis were explored. The best formulation provided parameter estimates with tight bounds on estimated covariances, suggesting that the model provides a reasonable description of the hydrogeologic system. Our investigation indicates the geometry of the recirculation zone and the degree of recirculation under two different sets of operating conditions. Surprisingly, our analysis suggests that the effects of aquifer heterogeneity are not significant at this site under the conditions of forced recirculation. Furthermore, anomalous flow through an open monitoring well created significant vertical short-circuiting between the generally insulated aquifers. Flow through this small open conduit was equivalent to as much as 33% of the flow through the pumping wells. Using the model as a guide, we treated the aquifer system and bioactive zones as an equivalent mixed reactor to develop simple expressions relating effluent concentrations to influent concentrations. We demonstrate how these expressions are useful in predicting the removal of TCE that had undergone in situ bioremediation in the recirculatory treatment well system. The finite element model developed in this work serves as the foundation for a reactive transport simulator that we developed to analyze bioremediation which occurred during a 444 day experiment <Gandhi et al., 2002>.