We study the possibility to employ neural networks to simulate jet clu
stering procedures in high energy hadron-hadron collisions. We concent
rate our analysis on the Fermilab Tevatron energy and on the k(perpend
icular to) algorithm. We employ both supervised and unsupervised neura
l networks. In the first case we consider a multilayer feed-forward ne
twork trained by the backpropagation algorithm: our results show that
these networks can satisfactorily simulate the relevant features of th
e k(perpendicular to) algorithm. We consider also unsupervised learnin
g, where the neural network autonomously organizes the events in clust
ers. The results of this analysis are discussed and compared with the
supervised approach.