In analog integrated circuits, process variations result in physical parameter variations. Simulated performance values must then be considered with their tolerance intervals. Consequently, contrarily to digital circuits where the outputs are either '0' or '1' such that we can decide without ambiguity whether a fault is detectable or not, for analog circuits fault detectability is still a value problem since the fault can either be completely detectable, partially detectable or completely undetectable which makes it very difficult to take a decision. In order to solve this decision problem, we have introduced the fault detection probability (FDP) function which allows to formalize the problem of analog fault detection subjected to parameter variations.