A formal theory of latent inhibition (LI) is offered in the context of
a real-time, neural network model of classical conditioning. The netw
ork assumes that the effectiveness of a conditioned stimulus (CS) in e
stablishing associations with the unconditioned stimulus (US) is propo
rtional to total novelty, defined as the sum of the absolute value of
the difference between the predicted and observed amplitudes of all en
vironmental events. CS effectiveness controls both the rate of storage
(formation, or read-in) and the retrieval (activation, or read-out) o
f CS-CS and CS-US associations. The model describes LI because total n
ovelty and, therefore, CS effectiveness decrease during CS preexposure
. Computer simulations demonstrate that the neural network correctly d
escribes, and sometimes predicts, the effects on LI of experimental ma
nipulations before and during CS preexposure and during and after cond
itioning.