In order to estimate a user’s head pose at a relative large scale environment for virtual reality (VR) applications, multiple cameras set around him/her are used in conventional approaches, such as a motion capture. This paper proposes a method of estimating head pose from spherical images. A user wears a helmet on which a visual sensor is mounted and the head pose can be estimated by observing the fiducial markers put around him/her. Since a spherical image has a full view, our method can cope with a big head rotation motion compared with a normal camera. Since a head pose at every time is directly estimated from the observed markers, there is no accumulated errors in our method compared with a inertial sensor. Currently, an omnidirectional image sensor is used to acquire the most part of a spherical image in our experiment.