A guaranteed cost robust centralized fusion steady-state Kalman predictor is presented for the multisensor system with uncertain noise variances based on the minimax robust estimation principle. A maximal perturbation region of uncertain noise covariances is obtained by using the Lagrange multiplier method. For all admissible perturbations in this region, the deviations of its actual accuracies with respect to the robust accuracy are guaranteed to remain within the prescribed range, and the maximal lower bound and minimal upper bound of accuracy deviations are given. The proof of the guaranteed cost is presented by using the Lyapunov equation approach. A simulation example is given to illustrate the correctness and effectiveness of the proposed method.