The purpose of the present experiment was to determine the preferred visual 'straight ahead' or anterior/posterior (a/p) axis at the perceptual level. The ability of 12 neurologically normal, young adult subjects to position a rod parallel to the head and trunk a/p axes while viewing eccentrically located visual targets were studied under six conditions: 1. fixed - subjects stood erect with the head aligned to the trunk and viewed a central target while visually aligning a hand-held rod to the head and trunk a/p axis. 2. eyes - subjects moved only their eyes to view eccentric targets and aligned the rod to the head and trunk a/p axis. 3. head-trunk - subjects viewed the eccentric targets by rotating the head about a vertical axis and aligned the rod to the trunk a/p axis. 4. head-head - subjects viewed the targets as in 3 and positioned the rod parallel to the head a/p axis. 5. trunk-head - subjects viewed the targets by rotating the trunk and head as a unit about the vertical axis and aligned the rod parallel to the head a/p axis (note that the head and trunk a/p axes were misaligned by the experimenter prior to target viewing). 6. trunk-trunk-subjects viewed targets as in 5 and positioned the rod parallel to the trunk a/p axis. Subjects performed 25-35 consecutive trials within each condition. Perceptual errors were similar for aligning the rod to the trunk and head a/p axes; however, moving the trunk produced much larger constant and variable perceptual errors than moving the head. In a second experiment, four subjects controlled the position of a lighted rod held by a robot arm in complete darkness. They were instructed to align the rod to either the head or trunk a/p axis under conditions similar to the fixed, head-trunk, and head-head tasks described above. Perceptual errors were much larger when aligning the rod to the head a/p axis than to trunk a/p axis when the head was moved. This shows that the trunk a/p axis is clearly preferred at the perceptual level when visual background cues are not present. These data strongly suggest that the visual coordinate system uses a trunk-fixed a/p axis to define the subjective straight-ahead direction and right/left position of a target. Implications of these findings for sensorimotor transformations in control of upper limb movements to visual targets are discussed.