This paper presents a scheme for fault detection and isolation (FDI) of on-board gyroscope sensors and thrusters for spacecraft attitude control, based on the example of the Mars Express (MEX) satellite. The main contribution of the paper is related to the design and the optimization of an FDI procedure based on robust observers or filters, used as estimators, which generate the FDI residual signals. When organized into an estimator bank, excellent fault isolation properties are achieved upon suitable design. The residual evaluation relies on decision logic, whose thresholds are properly selected and specified. The FDI strategy is applied to the non-linear simulation of the MEX system, and the FDI performance is evaluated subject to disturbance signals, model uncertainty, and measurement noise processes. The robustness and reliability properties of the robust residual generators are investigated and verified in simulation by selecting suitable performance criteria together with Monte Carlo analysis. The results obtained highlight a good trade-off between solution complexity and achieved performances. Comparisons with the existing fault diagnosis algorithms implemented on-board the MEX spacecraft are finally reported. The proposed FDI design methodology constitutes a reliability approach for real application of FDI in future spacecraft.