We have investigated how the paths of reaching movements improve with
motor learning, and whether these improvements transfer to movements o
ther than those in which subjects were trained. Planar reaching moveme
nts were recorded in three groups moving in diagonal and lateral direc
tions using a digitising table. All subjects made a number of reaching
movements in a pre-test session. In the subsequent training phase of
the experiment, one group of subjects was instructed to make lateral m
ovements with as straight a path as possible; a second group made simi
lar lateral movements following a straight line marked on the table; w
hile a third group made diagonal movements, also following a marked li
ne. All three groups were then tested making lateral and diagonal move
ments, without the benefit of any marked lines. The straightness and v
ariability of movement paths were analysed to investigate improvements
in neural control following training. A significant group by directio
n interaction indicated that movement straightness improved locally fo
r the directions which were trained. Movement variability, in contrast
, improved equally for all directions of movement. The results are con
sistent with local learning of a neural inverse kinematics model used
in movement planning and global learning of a neural forward kinematic
s model used in movement execution.