In this paper, an observer-based adaptive threshold is developed as part of a fault diagnosis scheme to detect and isolate commonly occurring faults in a vehicle alternator system. Since the mathematical model of the alternator subsystem is quite involved and highly nonlinear; in order to simplify the diagnostic scheme, an equivalent linear time varying model based on the input-output behavior of the system is used for threshold equations derivation. A novel approach using Gaussian distribution to obtain the parameters of the system is investigated. The validity of the proposed diagnosis scheme is tested through simulation and the results are presented.