Considering actuator faults occur when a microsatellite runs on orbit, nonlinear learning observer (NLO)-based fault reconstruction for satellite attitude control systems is investigated. Combined with the advantages of iterative learning algorithm and recursive learning algorithm, a novel learning algorithm involving current and previous measurement output errors is first proposed such that the proposed NLO can estimate satellite attitude angles and attitude angular velocities and reconstruct actuator faults accurately and quickly. Further, the stability conditions of the proposed NLO are provided and detailed design of observer gain matrices is given using the linear matrix inequality technique. At last, the proposed approach is applied to reconstruct thruster faults in microsatellites, simulation results validate the effectiveness of the proposed fault reconstruction approach.