A novel and pragmatic methodology is presented for the design of fault detection and isolation systems by using parity relations in a fully stochastic setting. The approach is based on data obtained, either from the real process or an analytical model, through a statistically designed experiment. The inspiring concept is that design optimality can only be attained if detection/isolation performance and robustness are simultaneously addressed in an integrated framework. Advantages with respect to the state of the art are pointed out, both in terms of general applicability of the method and optimality of the design solution. An illustrative case-study concludes the paper.