In this paper, a sigma-point Kalman filter formulation for attitude estimation is derived using the modified Rodrigues parameters and real data of attitude sensors. The unscented Kalman filter algorithm is used for attitude estimation and the gyro-based model is considered for attitude propagation. In this study, the attitude of satellite is estimated using real data supplied by gyros, Earth sensors and Sun sensors that are on board of the CBERS-2 (China Brazil Earth Resources Satellite). The results show that even with a sparse set of measures, the UKF with MRP shows results similar to those obtained when directly uses the Euler angles.