Knowledge acquisition for the student model of intelligent tutoring sy
stems (ITSs) remains a difficult problem, partly because of the comple
xity associated with understanding both how people learn and how it is
best to tutor, much of which relates to metacognition and problem-sol
ving skills. The bottleneck associated with this area significantly in
creases the development times of ITSs. Neural networks have made a mar
ked impact in many artificial intelligence areas such as pattern recog
nition, speech learning, speech understanding, and hand-written charac
ter recognition. Neural networks are noted for their ability to handle
noisy and approximate data, to generalize over situations they have n
ot handled before, and to be represented in a way amenable to parallel
processing. In addition, they have the ability to learn, a characteri
stic which should prove very useful in the development of ITSs. In thi
s paper, we show that neural networks can address the knowledge acquis
ition bottleneck associated with the student model. We demonstrate tha
t incomplete knowledge obtained from the expert can be refined and exp
anded by a neural network to provide a more complete, and hence more a
ccurate, student model.