In this paper, a robust heading determination method is proposed for low-cost attitude and heading reference system (AHRS) aided by the global positioning system (GPS). As compared with the traditional GPS/SINS-integrated navigation-based heading determination method, in the proposed method, the heading information obtained from the GPS velocity outputs is first incorporated into the observation vector, which constructs a novel completed GPS/SINS integration framework and greatly improves the observability of azimuthal misalignment in the Kalman filter. Moreover, a multivariate constrained total least square (MCTLS) method is proposed and integrated into the completed integration framework to deal with the measurement error in both input and output data of GPS velocity measurement model, which improves the accuracy of the observed heading information and yields a robust heading estimation at each time instant. Simulation and experiment results demonstrate that the proposed robust heading determination method can outperform the related state-of-the-art methods for the GPS-aided attitude and heading reference system.