Abstract: If a fault's occurrence is thought of as a discrete event. the system suffered from faults could be regarded as a stochastic hybrid system. So fault diagnosis is transformed into the discrete states estimation problem for such a stochastic hybrid system. Fault diagnosis problem of nonlinear system in arbitrary noise environment is studied with this point of view. If assumed that models after faults' occurrence are known, the system subject to these faults can be modeled with a stochastic hybrid automaton. Then hybrid states in such a stochastic hybrid system are estimated by particle filtering-based algorithm. Consequently, fault diagnosis task is accomplished. The feasibility of proposed approach is shown by simulation results, and it is suitable in diagnosing some type faults.