Abstract: A method of failure detection,isolation and accomodation for engine sensors is presented.The method is based on engine component level nonlinear model and extended Kalman filter to obtain optimal estimation values of failed sensors and calculation references of the engine model.This makes engine model adapt to the actual engine perfectly and provides analytical redundancy for the failed sensors.In this case,the reliability of the engine control system is enhanced greatly.A numerical digital simulation of a given turboshaft engine has been accomplished.The results prove the effectiveness and practicability of the present method.