This paper presents a method for designing a diagnosable system. The nature of a (locally) diagnosable system is first studied. The theoretical result is then applied to determine the proper set of test points so that the circuit under design is perfectly diagnosable. Two examples, one for a linear system and the other for a nonlinear system, are given. Our simulation outcome shows that the desired diagnosable system may not result if the choice of test points is limited by some constraints.