Automating model calibration and production optimization is computationally demanding because of the intensive multiphase-flow-simulation runs that are needed to predict the response of real reservoirs under proposed changes in model inputs. Fast surrogate models have been proposed to speed up reservoir-response predictions without compromising accuracy. Surrogate models either are derived by preserving the physics of the involved processes (e.g., mass balance) to provide reliable long-range predictions or are developed solely on the basis of statistical input/output relations, in which case they can only provide short-range predictions because of the absence of the physical processes that govern the long-term behavior of the reservoir. We present an alternative approach that combines the advantages of both statistics-based and physics-based methods by reducing the flow predictions in complex 3D models to a 1D flow-network model. The existing injection/production wells in the original model form the nodes or vertices of the flow network. Each pair of wells (nodes) in the flow network is connected by use of a 1D numerical simulation model, resulting in a connected network of 1D grid-based simulation models. The coupling between the individual 1D flow models is enforced at the nodes where network edges intersect. The proposed flow-network model provides a useful and fast tool for characterizing interwell connectivity, estimating drainage volume between each pair of wells, and predicting reservoir production over an extended period of time for optimization purposes. The parameters of the flow-network model are estimated by a robust training approach to ensure that the network model reproduces the response of the full model under a wide range of development strategies. This step helps the network model to preserve its predictive power during optimization iterations when alternative development strategies are proposed and evaluated to find the solution. We demonstrate the effectiveness and applicability of the proposed flow-network model by use of two-phase waterflooding experiments.