Without visual references, non-pilots exposed to coordinated flight turns underestimate the bank angle, because of discordant information of the roll-angular displacement from the otoliths, consistently signaling vertical position, versus the semicircular canals, enabling detection of the displacement. Pilots may also use their ability to perceive the G load and knowledge of the relation between load and angle to assess the bank angle. Our aim was to investigate if the perception of bank angle can be improved by spatial-orientation training in a centrifuge. Sixteen pilots/pilot students assessed their roll tilt, in complete darkness, both during real coordinated flight turns and gondola centrifugation, at roll tilts of 30° and 60°. The experiments were repeated after a 3-wk period, during which 8 of the subjects performed 9 training sessions in the centrifuge, comprising feedback on roll angle vs G load, and on indicating requested angles. Before training, the subjects perceived in the aircraft and centrifuge respectively: 37(17)°; 38(14)° during 60° turns, and 19(12)°; 20(10)° during 30° turns. Training improved the perception of angle during the 60° [to 60(7)°; 55(10)°; p≤0.04], but not the 30° turns [21(10)°; 15(9)°; p≥0.30]; the improvement disappeared within two years after training. Angle assessments did not change in the untrained group. The results suggest that it is possible to, in a centrifuge, train a pilot's ability to perceive large, but not discrete-to-moderate roll-angular displacements. The transient training effect is attributable to improved capacity to perceive and translate G load into roll angle and/or to increased reliance on semicircular-canal signals.