In this paper, a new algorithm for satellite attitude determination with sun sensor and magnetic sensor is designed that improves the accuracy of satellite attitude determination subsystem and robust it against sensor fault. This subsystem includes sensors and attitude determination algorithm. In order to combine sensors data, we can utilize  centralized and decentralized algorithm. Decentralized algorithm in addition of advantages such as computational volume reducing, improving accuracy and faster convergence, has appropriate structure for fault detection and isolation. In this paper, adaptive federated extended kalman filter is used for satellite attitude estimation and sensor fault detection and isolation that in addition of improving attitude determination subsystem performance, robust it against sensor fault. The correctness of aforementioned contents are demonstrated in simulations.