The robust state estimation problem for uncertain neural network is studied in this paper. As the uncertainty of the parameter, the states of the estimator can’t be complete synchronous with the neural network, but asymptotically synchronous with errorbound is accessible. For given state estimator gain matrix, the error bound is derived. By using stable theory and linear matrix inequality approach, the design of the robust state estimator is also given in this paper. And the discussion of the estimate of the error bound is also presented. The simulation samples have proved the effectiveness of the conclusion.