The vestibular aqueduct (VA) and paravestibular canaliculus (PVC) were reconstructed using a computer-aided three-dimentional reconstruction system (SERSERS) from five series of the serial, histopathological sections of the human tempolar bones without alignment markers. In order to align the serial sections, more than ten sectional tissue structure images on a section such as the cochlea, VA, PVC, semicircular canals, ossicles and facial canal were compared with those on the consecutive section and the position was determined where the deviation of each sectional image pair would be the minimum in every direction. Repeating this procedure throughout the series of the sections, we could perform the overall alignment of the sections. As the standard viewing axis of reconstruction, we used the modiolus and constructed a triplet of images ; posteromedial view image viewing from posterior surface side of the pyramis parallel to the modiolus, superior view image viewing from anterior surface side of the pyramis perpendicularly to the modiolus, anteromedial view image viewing from the internal carotid artery side perpendiculaly to the modiolus. The triplet images of the cochlea, VA and PVC presented us their three-dimentional configuration and the spatial relationship among them. Superior view showed that the angle between the plane of the proximal portion of the VA and that of the basal turn of cochlea varied in a wide range from 53 degrees to 68 degrees. Anteromedial view showed that the angle between the plane of the distal portion of the VA and that of the posterior semicircular canal also varied in a wide range from 22 degrees to 49 degrees. In comparison with a medial view graphic reconstruction method developed by Egami and Sando (1984), computer-aided three-dimentional reconstruction gives us more precise configurations of the VA and PVC and their spatial relationships to the posterior semicircular canal.