Brain-Machine Interfaces (BMIs) are being researched controlling external devices such as robots and computers by measuring the cranial nerve activity of the operator. The brain activities evoked by visual stimuli have been studied intensively. However, few studies have considered a BMI that uses the brain activities evoked by auditory stimuli. This study investigated whether a person's direction of attention can be estimated using an event-related potential (ERP) generated by selective attention to an auditory stimulus. An auditory stimulus and an out-of-head sound localization system that can create an audio image outside the head that is presented through an earphone were used instead of a loudspeaker system. This system was experimentally evaluated by presenting the subject auditory cues from one of six directions while the subject directed his attention in one direction. An EEG response similar to an ERP was observed. The direction of attention was estimated using support vector machine with an accuracy of 89.2[%] on average for the three subjects. This suggests that a BMI system based on the estimated direction of attention can be developed by using out-of-head sound localization.