The paper deals with the design of optimal filters for fault detection and isolation (FDI) in dynamic systems. The proposed method, involving the generation of directional residuals derived from the system＇s input-output description model, facilitates the enhancement of robustness even in cases where it is impossible to decouple fault effect from system perturbations. The proposed solution is based on the optimization of a performance index in the frequency domain. The resulting detection filter is capable of maximizing the effect of faults while minimizing the influence of disturbances.