Prevention and remediation strategies for groundwater pollution can be successfully carried out if the location, concentration, and release history of contaminants can be accurately identified. This, however, presents a challenge due to complex groundwater systems. To address this issue, a simulation-optimization (S/O) model by integrating MODFLOW and MT3DMS into a shuffled complex evolution (SCE-UA) optimization algorithm was proposed; this coupled model can identify the unknown groundwater pollution source characteristics. Moreover, the Grids Traversal algorithm was used for automatically searching all possible combinations of pollution source location. The performance of the proposed S/O model was tested by three hypothetical scenarios with various combinations of mixed situations (i.e., single and multiple pollution source locations, known and unknown pollution source locations, steady-state flow and transient flow). The field measurement errors was additionally considered and analyzed. Our results showed that this proposed S/O model performed reasonably well. The identified locations and concentrations of contaminants fairly matched with the imposed inputs with average normalized deviations less than 1% after sufficient generations. We further assessed the impact of generation number on the performance of the S/O model. The performance could be significantly improved by increasing generation number, which yet resulted in a heavy computational burden. Furthermore, the proposed S/O model performed more efficiently and robustly than the traditionally used artificial neural network (ANN)-based model. This is due to the internal linkage of numerical simulation in the S/O model that promotes the data exchange from external files to programming variables. This new model allows for solving the source-identification problems considering complex conditions, and thus for providing a platform for groundwater pollution prevention and management.