Suitable installed sensors in a industrial process is a necessary condition for fault diagnosis. Sensor placement for diagnosis purposes is to study which process variables have to be measured to satisfy diagnosis specifications (detectability, dis-criminability and diagnosability). This paper presents a method based on the study of the structural model properties and the Dulmage-Mendelsohn decomposition. Due to the use of structural models, the proposed approach can be applied to a wide variety of system (linear, algebraic, dynamics, etc.). Assuming that the cost of placing a sensor for each possible variable is defined, this method finds the minimal cost sensor configuration according to the diagnosability criteria. This method does not require the computation of testable subsystems