Automatic classification of flight paths related to observed aircraft noises is useful for unattended aircraft noise monitoring. This paper describes a method of detecting flight paths based on temporal changes in sound arrival direction. The method applies the acoustical cross-correlation technique using four microphones. The sound arrival direction is expressed as a set of unit vectors on the surface of a unit sphere assumed to be at the center of the microphones. A field experiment was carried out near an end of a runway to confirm the effectiveness of the procedure. The rate of correct classification was 93%.