In order to elucidate the role of the ''starting point'' in path integ
ration, normal subjects underwent a self-rotational task in a motor-dr
iven turntable rotating around an earth-vertical axis. They were passi
vely rotated (''stimulus'') and had to return to the starting point, c
ontrolling the direction and velocity of the turntable by means of a j
oystick (''response''). The test included conditions with an earth-fix
ed target (EFT) as starting point, shown before the stimulus, and cond
itions without EFT presentation. The subject's response always took pl
ace in total darkness. Subjects succeeded in returning to the starting
point in ail conditions but were more precise (i.e. had smaller varia
bility of responses) with the EFT than in the other conditions. The la
rger data scatter (inaccuracy) in these latter tasks was directly rela
ted to the return peak velocity, whereas with EFT there was no relatio
nship between amplitude and velocity of the return motion. These resul
ts suggest that the presentation of the starting point (the EFT) allow
s a real time integration to take place, thereby improving accuracy du
ring self-controlled motion in the dark. Five subjects were also teste
d with the same rotational paradigm in total darkness throughout, but
with the head in a different position during stimulus and response mot
ions. Thus, motion detection was performed by different semicircular c
anals during stimulus and response. The conditions used were head upri
ght during stimulus, hyper-extended backward during response motion, a
nd head backward during stimulus and upright during response motion. I
t was found that the accuracy during these tasks did not differ from t
hat during stimulus/response motion without change in the upright or b
ackward head position. These data indicate that estimates of trajector
y are, within limits, independent of canal plane.