A critical step in groundwater remediation is to identify locations and release histories of contaminant sources so that a cost-effective remediation strategy can be designed and so that clean-up costs can be partitioned between potential responsible parties. In this study, a contaminant source identification toolbox (CONSID) was developed to solve the source identification problem using MODFLOW and MT3DMS. It is assumed that the system of equations is uncertain due to the presence of both model uncertainty and measurement errors. Existing approaches for contaminant source identification seldom consider the impact of model uncertainty, which can be caused by model structure error, model parameter error, and numerical error. CONSID includes a specially developed constrained robust least squares (CRLS) estimator for solving uncertain systems in a least squares sense. CRLS directly incorporates prior knowledge about model uncertainty and measurement error and uses this information to determine a regularization parameter that is optimal for robustness. When the observation network is imperfect and the quantity and quality of observed data are poor, the resulting system can be illconditioned and the classic estimators may fail. CRLS becomes most useful in obtaining estimates in such situations. CRLS does not require probability distributions of model parameters; instead, it assumes that the worst-case scenario can be identified and the associated worst system perturbation is bounded. The effectiveness of the framework for contaminant source recovery has been demonstrated through both synthetic examples and real-case studies. Results show that CRLS achieves better performance than some other estimators for ill-conditioned uncertain systems.