This paper proposes the prediction error observer (PEOB) in the design of networked predictive control systems (NPCSs) with network delays and model errors and proposes stability criteria for closed-loop NPCSs. In the NPCSs, a prediction error occurs due to model errors. Based on a-priori predictions, the PEOB compensates for the prediction error when the controller predicts the current and future states. The proposed method improves the performance of the conventional NPC associated with network delays and model errors. Simulation results illustrate the effectiveness of the proposed control strategies.