A new neural representation for crossbar switching using Hopfield networks is presented. Under the theory of the asymptotical stability and instability of feasible solutions and infeasible solutions explored by Matsuda, the efficiency and safety of the network generated from this representation is theoretically guaranteed. That is, we clarify the worst quality feasible solutions the network may obtain and show that the network never obtains infeasible solutions. Thus, the unreliability in the behavior of neural systems, which is one of the most serious defects of neural systems for practical use, is overcome, and system performance is theoretically guaranteed. Furthermore, by clarifying the system performance of two other familiar Hopfield networks for crossbar switching that have already been proposed, the superiority of our network to those networks is theoretically shown in terms of practicality, reliability, efficiency, and safety. © 1998 Scripta Technica, Syst Comp Jpn 29(7): 1–11, 1998