An online fault detecting and isolating (FDI) method was proposed for f ault diagnosis of Infrared Earth Sensor. The Elman neural network was employed d ue to its capability of processing timevarying signals in real time. For the s a ke of simplicity, two Elman neural networks were developed for fault detecting ( FD) and fault isolating (FI) respectively. For FD, an FD Elman neural network an d corresponding logic judgment were designed to identify normal and faulty state s; the output of FD was the moment when a fault occurred. For FI, a novel gradie nt updating strategy was introduced in FI Elman neural network which does faulty pattern matching and conducts FI. Simulation results demonstrate that the FDI s trategy is real time, convergent available for output coupling, general with tim evarying signals. The proposed FDI can avoid modeling, so it is suitable for o nline FDI of satellite attitude control system (SACS).