In this paper, the matching of center lines from stereo tree images is con- sidered, and formulated as a combinatorial optimization problem. Marr and Poggio showed that the Hopfield network, a kind of artificial neural network model, can be applied to reconstruct 3D structure from random dot stereogram. Their method con- sists of the following five steps; 1.form matching candidates, 2.allocate candidates on units of network, 3.determine connection weights among units from the characteristics of images, 4.converge to a stable state and 5.reconstruct 3D structure from the state of network. This paper proposes a new formation of connection weights in the stereo line matching algorithm.