This paper describes an alignment algorithm that estimates the initial heading angle of AUVs (Autonomous Underwater Vehicle) for starting navigation in a sea area. In the basic dead reckoning system, the initial orientation of the vehicle is very important. In particular, the initial heading value is an essential factor in determining the performance of the entire navigation system. However, the heading angle of AUVs cannot be measured accurately because the DCS (Digital Compass) corrupted by surrounding magnetic field in pointing true north direction of the absolute global coordinate system (not the same to magnetic north direction). Therefore, we constructed an experimental constraint and designed an algorithm based on extended Kalman filter using only inertial navigation sensors and a GPS (Global Positioning System) receiver basically. The value of sensor covariance was selected by comparing the navigation results with the reference data. The proposed filter estimates the initial heading angle of AUVs for navigation in a sea area and reflects sampling characteristics of each sensor. Finally, we verify the performance of the filter through experiments.