The objective of this paper is to present a distributed model of the s
patiotemporal neural processing that underlies the control of two-dime
nsional saccadic eye movements in the monkey. In this new model the su
perior colliculus (SC) is represented by two layers of cells. Simulate
d visual inputs activate the upper layer that is connected lo the lowe
r (motor) layer with feed forward projections. Extensive lateral inter
connections exist in the motor layer The weights assigned to these int
erconnections are established with a recurrent back propagation algori
thm and by training on a set of activity patterns obtained from neuron
s recorded in the monkey SC. A distributed set of connections to horiz
ontal and vertical brainstem saccadic burst generators is trained to a
llow the model to make accurate saccades to randomly selected target p
ositions. Finally, the model is able to produce accurate eye movements
and realistic neural discharge for saccades evoked with a variety of
experimental conditions not included in the training set (for example,
averaging and express saccades).