In this paper, a robust stationary Kalman filter is designed by minimizing selected performance indices so that it is less sensitive to uncertainties. The uncertainties include not only stochastic factors such as process noise and measurement noise, but also deterministic factors such as unknown initial estimation error, modeling error and sensing bias. To reduce the effect on the uncertainties, three performance indices that should be minimized are selected based on the quantitative error analysis to both the deterministic and the stochastic uncertainties. The selected indices are the size of the observer gain, the condition number of the observer matrix, and the estimation error variance. The observer gain is obtained by optimally solving the multi-objectives optimization problem that minimizes the indices. The robustness of the proposed filter is demonstrated through the comparison with the standard Kalman filter.