This paper presents a scheme for satellite multi-sensor fault-tolerant attitude estimation. It can both detect a faulty sensor online and conduct fault tolerance in time. First, a satellite attitude estimator based on error states is presented as the unified filter algorithm, in which the filter state equation is built with the gyro model and satellite kinematics, and the measurement equations are built with angle sensors. Meanwhile, typical fault models of the angle sensors are given. Then, a fault-tolerant federated Kalman filter (FTFKF) scheme is designed. Its three sub-filters include the respective angle sensors and share the public gyro measurement. Under the sensor fault condition, a FTFKF can detect the faulty sensor by contrasting and analysing the dimensionless fault detection factors and then selectively fusing the sub-filters’ outputs to enable the satellite attitude determination accuracy to approach normal. The comparative analysis of simulation results in typical fault cases shows that the proposed satellite multi-sensor fault-tolerant attitude estimation scheme may achieve the expected fault-tolerant performance. It is promising for enhancing the reliability of satellite attitude determination and control systems.