Fault detection in micro-electrostatic actuators caused primarily by their mechanical components (combs, thin air-damping) is investigated in this article. The system is assumed to be linearly parameterizable and the parameter vector contains the quantities that are susceptible to faults. While the system is operating in a closed-loop configuration, a set-membership identifier monitors the feasible region within which the nominal parameters should reside. The hypervolume of this region is a measure of the uncertainty of the system parameters and decreases in a monotonie manner with time. A fault is detected when: a) there is a sudden increase in this volume, or b) when the identified centroid of the parameter vector resides out of the feasible region, or c) when the system's output is out of its allowable predicted bounds. Simulation studies are offered to test the efficiency of the suggested fault-detection method.