Network delays are inevitable in networked control systems and have negative effects on system performance and stability. To compensate network delay actively, a novel networked predictive control scheme based on a state observer is proposed in this paper. A state observer reconstructs the current states using the delayed output and gains of the state observer are designed using LMI methods. Stability conditions of the closed-loop networked predictive control system are derived. Using these conditions, less conservative upper bound of delay, which guarantees the asymptotic stability of the system, is obtained. The evaluation illustrates the effectiveness of the control strategy.