In this paper, we propose a novel approach to fault simulation of large mixed signal circuits using circuit partitioning and fault ordering. The conventional statistical fault model is divided, to separately account for global and local variations. We consider the problem of estimating the effect of a parameter deviation in a sub-module, on the system level specifications for a general non-linear circuit. This method uses a function approximation model to estimate the system response and rank the faults according to the severity. Applications of this algorithm include estimation of fault coverage and forms a key element in test generation and diagnosis procedures.